0% found this document useful (0 votes)

3K views1,183 pages

NumPy 2.0 Reference Guide

This document is the NumPy reference manual, which provides documentation on NumPy, a Python library used for working with arrays and matrices. It describes NumPy's N-dimensional array (ndarray) object and other key aspects like data types, indexing, universal functions (ufuncs), routines for array creation, manipulation, linear algebra, Fourier transforms, random number generation and more. The reference was written by the NumPy community and released on March 19, 2011.

Uploaded by

Lucian Ghitu

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

3K views1,183 pages

NumPy 2.0 Reference Guide

Uploaded by

Lucian Ghitu

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 1183

NumPy Reference

Release 2.0.0.dev-ac3cba3

Written by the NumPy community

March 19, 2011

CONTENTS

1 Array objects 3
1.1 The N-dimensional array (ndarray) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Scalars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
1.3 Data type objects (dtype) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
1.4 Indexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
1.5 Standard array subclasses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
1.6 Masked arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
1.7 The Array Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439

2 Universal functions (ufunc) 445

2.1 Broadcasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
2.2 Output type determination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
2.3 Use of internal buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
2.4 Error handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
2.5 Casting Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
2.6 ufunc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
2.7 Available ufuncs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458

3 Routines 463
3.1 Array creation routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
3.2 Array manipulation routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
3.3 Indexing routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
3.4 Data type routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555
3.5 Input and output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
3.6 Discrete Fourier Transform (numpy.fft) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589
3.7 Linear algebra (numpy.linalg) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 609
3.8 Random sampling (numpy.random) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 641
3.9 Sorting, searching, and counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
3.10 Logic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704
3.11 Binary operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 720
3.12 Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 727
3.13 Mathematical functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 747
3.14 Functional programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 807
3.15 Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 812
3.16 Financial functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824
3.17 Set routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 833
3.18 Window functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 837
3.19 Floating point error handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 849
3.20 Masked array operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855

i
3.21 Numpy-specific help functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 976
3.22 Miscellaneous routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 978
3.23 Test Support (numpy.testing) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 979
3.24 Asserts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 980
3.25 Mathematical functions with automatic domain (numpy.emath) . . . . . . . . . . . . . . . . . . . 990
3.26 Matrix library (numpy.matlib) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 990
3.27 Optionally Scipy-accelerated routines (numpy.dual) . . . . . . . . . . . . . . . . . . . . . . . . . 990
3.28 Numarray compatibility (numpy.numarray) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 991
3.29 Old Numeric compatibility (numpy.oldnumeric) . . . . . . . . . . . . . . . . . . . . . . . . . . 991
3.30 C-Types Foreign Function Interface (numpy.ctypeslib) . . . . . . . . . . . . . . . . . . . . . . 991
3.31 String operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 992

4 Packaging (numpy.distutils) 1027

4.1 Modules in numpy.distutils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1027
4.2 Building Installable C libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1038
4.3 Conversion of .src files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1039

5 Numpy C-API 1041

5.1 Python Types and C-Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1041
5.2 System configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1055
5.3 Data Type API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1057
5.4 Array API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1059
5.5 Array Iterator API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1095
5.6 UFunc API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1110
5.7 Generalized Universal Function API . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1116
5.8 Numpy core libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1118

6 Numpy internals 1125

6.1 Numpy C Code Explanations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1125
6.2 Internal organization of numpy arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1132
6.3 Multidimensional Array Indexing Order Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1133

7 Numpy and SWIG 1135

7.1 Numpy.i: a SWIG Interface File for NumPy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1135
7.2 Testing the numpy.i Typemaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1148

8 Acknowledgements 1151

Bibliography 1153

Python Module Index 1159

Index 1161

ii
NumPy Reference, Release 2.0.0.dev-ac3cba3

Release
2.0.dev-ac3cba3
Date
March 19, 2011
This reference manual details functions, modules, and objects included in Numpy, describing what they are and what
they do. For learning how to use NumPy, see also user.

CONTENTS 1
NumPy Reference, Release 2.0.0.dev-ac3cba3

2 CONTENTS
CHAPTER

ONE

ARRAY OBJECTS

NumPy provides an N-dimensional array type, the ndarray, which describes a collection of “items” of the same type.
The items can be indexed using for example N integers.
All ndarrays are homogenous: every item takes up the same size block of memory, and all blocks are interpreted in
exactly the same way. How each item in the array is to be interpreted is specified by a separate data-type object, one
of which is associated with every array. In addition to basic types (integers, floats, etc.), the data type objects can also
represent data structures.
An item extracted from an array, e.g., by indexing, is represented by a Python object whose type is one of the array
scalar types built in Numpy. The array scalars allow easy manipulation of also more complicated arrangements of
data.

Figure 1.1: Figure Conceptual diagram showing the relationship between the three fundamental objects used to de-
scribe the data in an array: 1) the ndarray itself, 2) the data-type object that describes the layout of a single fixed-size
element of the array, 3) the array-scalar Python object that is returned when a single element of the array is accessed.

1.1 The N-dimensional array (ndarray)

An ndarray is a (usually fixed-size) multidimensional container of items of the same type and size. The number
of dimensions and items in an array is defined by its shape, which is a tuple of N positive integers that specify
the sizes of each dimension. The type of items in the array is specified by a separate data-type object (dtype), one of
which is associated with each ndarray.
As with other container objects in Python, the contents of an ndarray can be accessed and modified by indexing
or slicing the array (using, for example, N integers), and via the methods and attributes of the ndarray. Different

3
NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarrays can share the same data, so that changes made in one ndarray may be visible in another. That is, an
ndarray can be a “view” to another ndarray, and the data it is referring to is taken care of by the “base” ndarray.
ndarrays can also be views to memory owned by Python strings or objects implementing the buffer or array
interfaces.
Example
A 2-dimensional array of size 2 x 3, composed of 4-byte integer elements:
>>> x = np.array([[1, 2, 3], [4, 5, 6]], np.int32)
>>> type(x)
<type ’numpy.ndarray’>
>>> x.shape
(2, 3)
>>> x.dtype
dtype(’int32’)

The array can be indexed using Python container-like syntax:

>>> x[1,2] # i.e., the element of x in the *second* row, *third*
column, namely, 6.

For example slicing can produce views of the array:

>>> y = x[:,1]
>>> y
array([2, 5])
>>> y[0] = 9 # this also changes the corresponding element in x
>>> y
array([9, 5])
>>> x
array([[1, 9, 3],
[4, 5, 6]])

1.1.1 Constructing arrays

New arrays can be constructed using the routines detailed in Array creation routines, and also by using the low-level
ndarray constructor:
ndarray An array object represents a multidimensional, homogeneous array of fixed-size items.
class numpy.ndarray
An array object represents a multidimensional, homogeneous array of fixed-size items. An associated data-type
object describes the format of each element in the array (its byte-order, how many bytes it occupies in memory,
whether it is an integer, a floating point number, or something else, etc.)
Arrays should be constructed using array, zeros or empty (refer to the See Also section below). The parameters
given here refer to a low-level method (ndarray(...)) for instantiating an array.
For more information, refer to the numpy module and examine the the methods and attributes of an array.
Parameters
(for the __new__ method; see Notes below) :
shape : tuple of ints
Shape of created array.
dtype : data-type, optional
Any object that can be interpreted as a numpy data type.

4 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

buffer : object exposing buffer interface, optional

Used to fill the array with data.
offset : int, optional
Offset of array data in buffer.
strides : tuple of ints, optional
Strides of data in memory.
order : {‘C’, ‘F’}, optional
Row-major or column-major order.
See Also:

array
Construct an array.
zeros
Create an array, each element of which is zero.
empty
Create an array, but leave its allocated memory unchanged (i.e., it contains “garbage”).
dtype
Create a data-type.

Notes
There are two modes of creating an array using __new__:
1.If buffer is None, then only shape, dtype, and order are used.
2.If buffer is an object exposing the buffer interface, then all keywords are interpreted.
No __init__ method is needed because the array is fully initialized after the __new__ method.

Examples
These examples illustrate the low-level ndarray constructor. Refer to the See Also section above for easier ways
of constructing an ndarray.
First mode, buffer is None:
>>> np.ndarray(shape=(2,2), dtype=float, order=’F’)
array([[ -1.13698227e+002, 4.25087011e-303],
[ 2.88528414e-306, 3.27025015e-309]]) #random

Second mode:
>>> np.ndarray((2,), buffer=np.array([1,2,3]),
... offset=np.int_().itemsize,
... dtype=int) # offset = 1*itemsize, i.e. skip first element
array([2, 3])

1.1. The N-dimensional array (ndarray) 5

NumPy Reference, Release 2.0.0.dev-ac3cba3

Attributes
T
data
dtype Create a data type object.
flags
flat
imag(val) Return the imaginary part of the elements of the array.
real(val) Return the real part of the elements of the array.
size(a[, axis]) Return the number of elements along a given axis.
itemsize
nbytes Base object for a dictionary for look-up with any alias for an array dtype.
ndim(a) Return the number of dimensions of an array.
shape(a) Return the shape of an array.
strides
ctypes
base
class numpy.dtype
Create a data type object.
A numpy array is homogeneous, and contains elements described by a dtype object. A dtype object can be
constructed from different combinations of fundamental numeric types.
Parameters
obj :
Object to be converted to a data type object.
align : bool, optional
Add padding to the fields to match what a C compiler would output for a similar C-
struct. Can be True only if obj is a dictionary or a comma-separated string.
copy : bool, optional
Make a new copy of the data-type object. If False, the result may just be a reference
to a built-in data-type object.

Examples
Using array-scalar type:
>>> np.dtype(np.int16)
dtype(’int16’)

Record, one field name ‘f1’, containing int16:

>>> np.dtype([(’f1’, np.int16)])
dtype([(’f1’, ’<i2’)])

Record, one field named ‘f1’, in itself containing a record with one field:
>>> np.dtype([(’f1’, [(’f1’, np.int16)])])
dtype([(’f1’, [(’f1’, ’<i2’)])])

Record, two fields: the first field contains an unsigned int, the second an int32:
>>> np.dtype([(’f1’, np.uint), (’f2’, np.int32)])
dtype([(’f1’, ’<u4’), (’f2’, ’<i4’)])

6 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Using array-protocol type strings:

>>> np.dtype([(’a’,’f8’),(’b’,’S10’)])
dtype([(’a’, ’<f8’), (’b’, ’|S10’)])

Using comma-separated field formats. The shape is (2,3):

>>> np.dtype("i4, (2,3)f8")
dtype([(’f0’, ’<i4’), (’f1’, ’<f8’, (2, 3))])

Using tuples. int is a fixed type, 3 the field’s shape. void is a flexible type, here of size 10:
>>> np.dtype([(’hello’,(np.int,3)),(’world’,np.void,10)])
dtype([(’hello’, ’<i4’, 3), (’world’, ’|V10’)])

Subdivide int16 into 2 int8‘s, called x and y. 0 and 1 are the offsets in bytes:
>>> np.dtype((np.int16, {’x’:(np.int8,0), ’y’:(np.int8,1)}))
dtype((’<i2’, [(’x’, ’|i1’), (’y’, ’|i1’)]))

Using dictionaries. Two fields named ‘gender’ and ‘age’:

>>> np.dtype({’names’:[’gender’,’age’], ’formats’:[’S1’,np.uint8]})
dtype([(’gender’, ’|S1’), (’age’, ’|u1’)])

Offsets in bytes, here 0 and 25:

>>> np.dtype({’surname’:(’S25’,0),’age’:(np.uint8,25)})
dtype([(’surname’, ’|S25’), (’age’, ’|u1’)])

Methods
newbyteorder
numpy.imag(val)
Return the imaginary part of the elements of the array.
Parameters
val : array_like
Input array.
Returns
out : ndarray
Output array. If val is real, the type of val is used for the output. If val has complex
elements, the returned type is float.
See Also:
real, angle, real_if_close

Examples
>>> a = np.array([1+2j, 3+4j, 5+6j])
>>> a.imag
array([ 2., 4., 6.])
>>> a.imag = np.array([8, 10, 12])
>>> a
array([ 1. +8.j, 3.+10.j, 5.+12.j])

numpy.real(val)
Return the real part of the elements of the array.

1.1. The N-dimensional array (ndarray) 7

NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
val : array_like
Input array.
Returns
out : ndarray
Output array. If val is real, the type of val is used for the output. If val has complex
elements, the returned type is float.
See Also:
real_if_close, imag, angle

Examples
>>> a = np.array([1+2j, 3+4j, 5+6j])
>>> a.real
array([ 1., 3., 5.])
>>> a.real = 9
>>> a
array([ 9.+2.j, 9.+4.j, 9.+6.j])
>>> a.real = np.array([9, 8, 7])
>>> a
array([ 9.+2.j, 8.+4.j, 7.+6.j])

Methods

8 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Table 1.1 – continued from previous page

newbyteorder
nonzero(a) Return the indices of the elements that are non-zero.
prod(a[, axis, dtype, out]) Return the product of array elements over a given axis.
ptp(a[, axis, out]) Range of values (maximum - minimum) along an axis.
put(a, ind, v[, mode]) Replaces specified elements of an array with given values.
ravel(a[, order]) Return a flattened array.
repeat(a, repeats[, axis]) Repeat elements of an array.
reshape(a, newshape[, order]) Gives a new shape to an array without changing its data.
resize(a, new_shape) Return a new array with the specified shape.
round(a[, decimals, out]) Round an array to the given number of decimals.
searchsorted(a, v[, side]) Find indices where elements should be inserted to maintain order.
setasflat
setfield
setflags
sort(a[, axis, kind, order]) Return a sorted copy of an array.
squeeze(a) Remove single-dimensional entries from the shape of an array.
std(a[, axis, dtype, out, ddof]) Compute the standard deviation along the specified axis.
sum(a[, axis, dtype, out]) Sum of array elements over a given axis.
swapaxes(a, axis1, axis2) Interchange two axes of an array.
take(a, indices[, axis, out, mode]) Take elements from an array along an axis.
tofile
tolist
tostring
trace(a[, offset, axis1, axis2, dtype, out]) Return the sum along diagonals of the array.
transpose(a[, axes]) Permute the dimensions of an array.
var(a[, axis, dtype, out, ddof]) Compute the variance along the specified axis.
view

numpy.all(a, axis=None, out=None)

Test whether all array elements along a given axis evaluate to True.
Parameters
a : array_like
Input array or object that can be converted to an array.
axis : int, optional
Axis along which a logical AND is performed. The default (axis = None) is to perform
a logical AND over a flattened input array. axis may be negative, in which case it counts
from the last to the first axis.
out : ndarray, optional
Alternate output array in which to place the result. It must have the same shape as
the expected output and its type is preserved (e.g., if dtype(out) is float, the result
will consist of 0.0’s and 1.0’s). See doc.ufuncs (Section “Output arguments”) for more
details.
Returns
all : ndarray, bool
A new boolean or array is returned unless out is specified, in which case a reference to
out is returned.
See Also:

1.1. The N-dimensional array (ndarray) 9

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.all
equivalent method
any
Test whether any element along a given axis evaluates to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.all([[True,False],[True,True]])
False

>>> np.all([[True,False],[True,True]], axis=0)

array([ True, False], dtype=bool)

>>> np.all([-1, 4, 5])

True

>>> np.all([1.0, np.nan])

True

>>> o=np.array([False])
>>> z=np.all([-1, 4, 5], out=o)
>>> id(z), id(o), z
(28293632, 28293632, array([ True], dtype=bool))

numpy.any(a, axis=None, out=None)

Test whether any array element along a given axis evaluates to True.
Returns single boolean unless axis is not None
Parameters
a : array_like
Input array or object that can be converted to an array.
axis : int, optional
Axis along which a logical OR is performed. The default (axis = None) is to perform
a logical OR over a flattened input array. axis may be negative, in which case it counts
from the last to the first axis.
out : ndarray, optional
Alternate output array in which to place the result. It must have the same shape as the
expected output and its type is preserved (e.g., if it is of type float, then it will remain
so, returning 1.0 for True and 0.0 for False, regardless of the type of a). See doc.ufuncs
(Section “Output arguments”) for details.
Returns
any : bool or ndarray
A new boolean or ndarray is returned unless out is specified, in which case a reference
to out is returned.
See Also:

10 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.any
equivalent method
all
Test whether all elements along a given axis evaluate to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.any([[True, False], [True, True]])
True

>>> np.any([[True, False], [False, False]], axis=0)

array([ True, False], dtype=bool)

>>> np.any([-1, 0, 5])

True

>>> np.any(np.nan)
True

>>> o=np.array([False])
>>> z=np.any([-1, 4, 5], out=o)
>>> z, o
(array([ True], dtype=bool), array([ True], dtype=bool))
>>> # Check now that z is a reference to o
>>> z is o
True
>>> id(z), id(o) # identity of z and o
(191614240, 191614240)

numpy.argmax(a, axis=None)
Indices of the maximum values along an axis.
Parameters
a : array_like
Input array.
axis : int, optional
By default, the index is into the flattened array, otherwise along the specified axis.
Returns
index_array : ndarray of ints
Array of indices into the array. It has the same shape as a.shape with the dimension
along axis removed.
See Also:
ndarray.argmax, argmin
amax
The maximum value along a given axis.
unravel_index
Convert a flat index into an index tuple.

1.1. The N-dimensional array (ndarray) 11

NumPy Reference, Release 2.0.0.dev-ac3cba3

Notes
In case of multiple occurrences of the maximum values, the indices corresponding to the first occurrence
are returned.

Examples
>>> a = np.arange(6).reshape(2,3)
>>> a
array([[0, 1, 2],
[3, 4, 5]])
>>> np.argmax(a)
5
>>> np.argmax(a, axis=0)
array([1, 1, 1])
>>> np.argmax(a, axis=1)
array([2, 2])

>>> b = np.arange(6)
>>> b[1] = 5
>>> b
array([0, 5, 2, 3, 4, 5])
>>> np.argmax(b) # Only the first occurrence is returned.
1

numpy.argmin(a, axis=None)
Return the indices of the minimum values along an axis.
See Also:

argmax
Similar function. Please refer to numpy.argmax for detailed documentation.

numpy.argsort(a, axis=-1, kind=’quicksort’, order=None)

Returns the indices that would sort an array.
Perform an indirect sort along the given axis using the algorithm specified by the kind keyword. It returns
an array of indices of the same shape as a that index data along the given axis in sorted order.
Parameters
a : array_like
Array to sort.
axis : int or None, optional
Axis along which to sort. The default is -1 (the last axis). If None, the flattened array is
used.
kind : {‘quicksort’, ‘mergesort’, ‘heapsort’}, optional
Sorting algorithm.
order : list, optional
When a is an array with fields defined, this argument specifies which fields to compare
first, second, etc. Not all fields need be specified.
Returns
index_array : ndarray, int
Array of indices that sort a along the specified axis. In other words,
a[index_array] yields a sorted a.

12 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:

sort
Describes sorting algorithms used.
lexsort
Indirect stable sort with multiple keys.
ndarray.sort
Inplace sort.

Notes
See sort for notes on the different sorting algorithms.
As of NumPy 1.4.0 argsort works with real/complex arrays containing nan values. The enhanced sort
order is documented in sort.

Examples
One dimensional array:
>>> x = np.array([3, 1, 2])
>>> np.argsort(x)
array([1, 2, 0])

Two-dimensional array:
>>> x = np.array([[0, 3], [2, 2]])
>>> x
array([[0, 3],
[2, 2]])

>>> np.argsort(x, axis=0)

array([[0, 1],
[1, 0]])

>>> np.argsort(x, axis=1)

array([[0, 1],
[0, 1]])

Sorting with keys:

>>> x = np.array([(1, 0), (0, 1)], dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])
>>> x
array([(1, 0), (0, 1)],
dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])

>>> np.argsort(x, order=(’x’,’y’))

array([1, 0])

>>> np.argsort(x, order=(’y’,’x’))

array([0, 1])

numpy.choose(a, choices, out=None, mode=’raise’)

1.1. The N-dimensional array (ndarray) 13

NumPy Reference, Release 2.0.0.dev-ac3cba3

np.choose(a,c) == np.array([c[a[I]][I] for I in ndi.ndindex(a.shape)]).

But this omits some subtleties. Here is a fully general summary:
Given an “index” array (a) of integers and a sequence of n arrays (choices), a and each choice array are
first broadcast, as necessary, to arrays of a common shape; calling these Ba and Bchoices[i], i = 0,...,n-1
we have that, necessarily, Ba.shape == Bchoices[i].shape for each i. Then, a new array with
shape Ba.shape is created as follows:
•if mode=raise (the default), then, first of all, each element of a (and thus Ba) must be in the range
[0, n-1]; now, suppose that i (in that range) is the value at the (j0, j1, ..., jm) position in Ba - then the
value at the same position in the new array is the value in Bchoices[i] at that same position;
•if mode=wrap, values in a (and thus Ba) may be any (signed) integer; modular arithmetic is used to
map integers outside the range [0, n-1] back into that range; and then the new array is constructed as
above;
•if mode=clip, values in a (and thus Ba) may be any (signed) integer; negative integers are mapped
to 0; values greater than n-1 are mapped to n-1; and then the new array is constructed as above.

Parameters
a : int array
This array must contain integers in [0, n-1], where n is the number of choices, unless
mode=wrap or mode=clip, in which cases any integers are permissible.
choices : sequence of arrays
Choice arrays. a and all of the choices must be broadcastable to the same shape. If
choices is itself an array (not recommended), then its outermost dimension (i.e., the one
corresponding to choices.shape[0]) is taken as defining the “sequence”.
out : array, optional
If provided, the result will be inserted into this array. It should be of the appropriate
shape and dtype.
mode : {‘raise’ (default), ‘wrap’, ‘clip’}, optional
Specifies how indices outside [0, n-1] will be treated:
• ‘raise’ : an exception is raised
• ‘wrap’ : value becomes value mod n
• ‘clip’ : values < 0 are mapped to 0, values > n-1 are mapped to n-1
Returns
merged_array : array
The merged result.
Raises
ValueError: shape mismatch :
If a and each choice array are not all broadcastable to the same shape.

See Also:

ndarray.choose
equivalent method

14 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Notes
To reduce the chance of misinterpretation, even though the following “abuse” is nominally supported,
choices should neither be, nor be thought of as, a single array, i.e., the outermost sequence-like container
should be either a list or a tuple.

Examples
>>> choices = [[0, 1, 2, 3], [10, 11, 12, 13],
... [20, 21, 22, 23], [30, 31, 32, 33]]
>>> np.choose([2, 3, 1, 0], choices
... # the first element of the result will be the first element of the
... # third (2+1) "array" in choices, namely, 20; the second element
... # will be the second element of the fourth (3+1) choice array, i.e.,
... # 31, etc.
... )
array([20, 31, 12, 3])
>>> np.choose([2, 4, 1, 0], choices, mode=’clip’) # 4 goes to 3 (4-1)
array([20, 31, 12, 3])
>>> # because there are 4 choice arrays
>>> np.choose([2, 4, 1, 0], choices, mode=’wrap’) # 4 goes to (4 mod 4)
array([20, 1, 12, 3])
>>> # i.e., 0

A couple examples illustrating how choose broadcasts:

>>> a = [[1, 0, 1], [0, 1, 0], [1, 0, 1]]
>>> choices = [-10, 10]
>>> np.choose(a, choices)
array([[ 10, -10, 10],
[-10, 10, -10],
[ 10, -10, 10]])

>>> # With thanks to Anne Archibald

>>> a = np.array([0, 1]).reshape((2,1,1))
>>> c1 = np.array([1, 2, 3]).reshape((1,3,1))
>>> c2 = np.array([-1, -2, -3, -4, -5]).reshape((1,1,5))
>>> np.choose(a, (c1, c2)) # result is 2x3x5, res[0,:,:]=c1, res[1,:,:]=c2
array([[[ 1, 1, 1, 1, 1],
[ 2, 2, 2, 2, 2],
[ 3, 3, 3, 3, 3]],
[[-1, -2, -3, -4, -5],
[-1, -2, -3, -4, -5],
[-1, -2, -3, -4, -5]]])

numpy.clip(a, a_min, a_max, out=None)

1.1. The N-dimensional array (ndarray) 15

NumPy Reference, Release 2.0.0.dev-ac3cba3

Maximum value. If a_min or a_max are array_like, then they will be broadcasted to the
shape of a.
out : ndarray, optional
The results will be placed in this array. It may be the input array for in-place clipping.
out must be of the right shape to hold the output. Its type is preserved.
Returns
clipped_array : ndarray
An array with the elements of a, but where values < a_min are replaced with a_min, and
those > a_max with a_max.
See Also:

numpy.doc.ufuncs
Section “Output arguments”

Examples
>>> a = np.arange(10)
>>> np.clip(a, 1, 8)
array([1, 1, 2, 3, 4, 5, 6, 7, 8, 8])
>>> a
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> np.clip(a, 3, 6, out=a)
array([3, 3, 3, 3, 4, 5, 6, 6, 6, 6])
>>> a = np.arange(10)
>>> a
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> np.clip(a, [3,4,1,1,1,4,4,4,4,4], 8)
array([3, 4, 2, 3, 4, 5, 6, 7, 8, 8])

numpy.compress(condition, a, axis=None, out=None)

Return selected slices of an array along given axis.
When working along a given axis, a slice along that axis is returned in output for each index where condi-
tion evaluates to True. When working on a 1-D array, compress is equivalent to extract.
Parameters
condition : 1-D array of bools
Array that selects which entries to return. If len(condition) is less than the size of a
along the given axis, then output is truncated to the length of the condition array.
a : array_like
Array from which to extract a part.
axis : int, optional
Axis along which to take slices. If None (default), work on the flattened array.
out : ndarray, optional
Output array. Its type is preserved and it must be of the right shape to hold the output.
Returns
compressed_array : ndarray
A copy of a without the slices along axis for which condition is false.

16 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:
take, choose, diag, diagonal, select
ndarray.compress
Equivalent method.
numpy.doc.ufuncs
Section “Output arguments”

Examples
>>> a = np.array([[1, 2], [3, 4], [5, 6]])
>>> a
array([[1, 2],
[3, 4],
[5, 6]])
>>> np.compress([0, 1], a, axis=0)
array([[3, 4]])
>>> np.compress([False, True, True], a, axis=0)
array([[3, 4],
[5, 6]])
>>> np.compress([False, True], a, axis=1)
array([[2],
[4],
[6]])

Working on the flattened array does not return slices along an axis but selects elements.
>>> np.compress([False, True], a)
array([2])

numpy.conj(x[, out ])
Return the complex conjugate, element-wise.
The complex conjugate of a complex number is obtained by changing the sign of its imaginary part.
Parameters
x : array_like
Input value.
Returns
y : ndarray
The complex conjugate of x, with same dtype as y.

Examples
>>> np.conjugate(1+2j)
(1-2j)

>>> x = np.eye(2) + 1j * np.eye(2)

>>> np.conjugate(x)
array([[ 1.-1.j, 0.-0.j],
[ 0.-0.j, 1.-1.j]])

numpy.copy(a)
Return an array copy of the given object.
Parameters
a : array_like

1.1. The N-dimensional array (ndarray) 17

NumPy Reference, Release 2.0.0.dev-ac3cba3

Input data.
Returns
arr : ndarray
Array interpretation of a.

Notes
This is equivalent to
>>> np.array(a, copy=True)

Examples
Create an array x, with a reference y and a copy z:
>>> x = np.array([1, 2, 3])
>>> y = x
>>> z = np.copy(x)

Note that, when we modify x, y changes, but not z:

>>> x[0] = 10
>>> x[0] == y[0]
True
>>> x[0] == z[0]
False

numpy.cumprod(a, axis=None, dtype=None, out=None)

Return the cumulative product of elements along a given axis.
Parameters
a : array_like
Input array.
axis : int, optional
Axis along which the cumulative product is computed. By default the input is flattened.
dtype : dtype, optional
Type of the returned array, as well as of the accumulator in which the elements are
multiplied. If dtype is not specified, it defaults to the dtype of a, unless a has an integer
dtype with a precision less than that of the default platform integer. In that case, the
default platform integer is used instead.
out : ndarray, optional
Alternative output array in which to place the result. It must have the same shape and
buffer length as the expected output but the type of the resulting values will be cast if
necessary.
Returns
cumprod : ndarray
A new array holding the result is returned unless out is specified, in which case a refer-
ence to out is returned.
See Also:

numpy.doc.ufuncs
Section “Output arguments”

18 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

Examples
>>> a = np.array([1,2,3])
>>> np.cumprod(a) # intermediate results 1, 1*2
... # total product 1*2*3 = 6
array([1, 2, 6])
>>> a = np.array([[1, 2, 3], [4, 5, 6]])
>>> np.cumprod(a, dtype=float) # specify type of output
array([ 1., 2., 6., 24., 120., 720.])

The cumulative product for each column (i.e., over the rows) of a:
>>> np.cumprod(a, axis=0)
array([[ 1, 2, 3],
[ 4, 10, 18]])

The cumulative product for each row (i.e. over the columns) of a:
>>> np.cumprod(a,axis=1)
array([[ 1, 2, 6],
[ 4, 20, 120]])

numpy.cumsum(a, axis=None, dtype=None, out=None)

Return the cumulative sum of the elements along a given axis.
Parameters
a : array_like
Input array.
axis : int, optional
Axis along which the cumulative sum is computed. The default (None) is to compute
the cumsum over the flattened array.
dtype : dtype, optional
Type of the returned array and of the accumulator in which the elements are summed. If
dtype is not specified, it defaults to the dtype of a, unless a has an integer dtype with a
precision less than that of the default platform integer. In that case, the default platform
integer is used.
out : ndarray, optional
Alternative output array in which to place the result. It must have the same shape and
buffer length as the expected output but the type will be cast if necessary. See doc.ufuncs
(Section “Output arguments”) for more details.
Returns
cumsum_along_axis : ndarray.
A new array holding the result is returned unless out is specified, in which case a ref-
erence to out is returned. The result has the same size as a, and the same shape as a if
axis is not None or a is a 1-d array.
See Also:

sum
Sum array elements.

1.1. The N-dimensional array (ndarray) 19

NumPy Reference, Release 2.0.0.dev-ac3cba3

trapz
Integration of array values using the composite trapezoidal rule.

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

Examples
>>> a = np.array([[1,2,3], [4,5,6]])
>>> a
array([[1, 2, 3],
[4, 5, 6]])
>>> np.cumsum(a)
array([ 1, 3, 6, 10, 15, 21])
>>> np.cumsum(a, dtype=float) # specifies type of output value(s)
array([ 1., 3., 6., 10., 15., 21.])

>>> np.cumsum(a,axis=0) # sum over rows for each of the 3 columns

array([[1, 2, 3],
[5, 7, 9]])
>>> np.cumsum(a,axis=1) # sum over columns for each of the 2 rows
array([[ 1, 3, 6],
[ 4, 9, 15]])

numpy.diagonal(a, offset=0, axis1=0, axis2=1)

Return specified diagonals.
If a is 2-D, returns the diagonal of a with the given offset, i.e., the collection of elements of the form
a[i, i+offset]. If a has more than two dimensions, then the axes specified by axis1 and axis2 are
used to determine the 2-D sub-array whose diagonal is returned. The shape of the resulting array can
be determined by removing axis1 and axis2 and appending an index to the right equal to the size of the
resulting diagonals.
Parameters
a : array_like
Array from which the diagonals are taken.
offset : int, optional
Offset of the diagonal from the main diagonal. Can be positive or negative. Defaults to
main diagonal (0).
axis1 : int, optional
Axis to be used as the first axis of the 2-D sub-arrays from which the diagonals should
be taken. Defaults to first axis (0).
axis2 : int, optional
Axis to be used as the second axis of the 2-D sub-arrays from which the diagonals
should be taken. Defaults to second axis (1).
Returns
array_of_diagonals : ndarray
If a is 2-D, a 1-D array containing the diagonal is returned. If the dimension of a is
larger, then an array of diagonals is returned, “packed” from left-most dimension to
right-most (e.g., if a is 3-D, then the diagonals are “packed” along rows).

20 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Raises
ValueError :
If the dimension of a is less than 2.
See Also:

diag
MATLAB work-a-like for 1-D and 2-D arrays.
diagflat
Create diagonal arrays.
trace
Sum along diagonals.

Examples
>>> a = np.arange(4).reshape(2,2)
>>> a
array([[0, 1],
[2, 3]])
>>> a.diagonal()
array([0, 3])
>>> a.diagonal(1)
array([1])

A 3-D example:
>>> a = np.arange(8).reshape(2,2,2); a
array([[[0, 1],
[2, 3]],
[[4, 5],
[6, 7]]])
>>> a.diagonal(0, # Main diagonals of two arrays created by skipping
... 0, # across the outer(left)-most axis last and
... 1) # the "middle" (row) axis first.
array([[0, 6],
[1, 7]])

The sub-arrays whose main diagonals we just obtained; note that each corresponds to fixing the right-most
(column) axis, and that the diagonals are “packed” in rows.
>>> a[:,:,0] # main diagonal is [0 6]
array([[0, 2],
[4, 6]])
>>> a[:,:,1] # main diagonal is [1 7]
array([[1, 3],
[5, 7]])

numpy.dot(a, b, out=None)
Dot product of two arrays.
For 2-D arrays it is equivalent to matrix multiplication, and for 1-D arrays to inner product of vectors
(without complex conjugation). For N dimensions it is a sum product over the last axis of a and the
second-to-last of b:
dot(a, b)[i,j,k,m] = sum(a[i,j,:] * b[k,:,m])

1.1. The N-dimensional array (ndarray) 21

NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
a : array_like
First argument.
b : array_like
Second argument.
out : ndarray, optional
Output argument. This must have the exact kind that would be returned if it was not
used. In particular, it must have the right type, must be C-contiguous, and its dtype
must be the dtype that would be returned for dot(a,b). This is a performance feature.
Therefore, if these conditions are not met, an exception is raised, instead of attempting
to be flexible.
Returns
output : ndarray
Returns the dot product of a and b. If a and b are both scalars or both 1-D arrays then a
scalar is returned; otherwise an array is returned. If out is given, then it is returned.
Raises
ValueError :
If the last dimension of a is not the same size as the second-to-last dimension of b.

See Also:

vdot
Complex-conjugating dot product.
tensordot
Sum products over arbitrary axes.
einsum
Einstein summation convention.

Examples
>>> np.dot(3, 4)
12

Neither argument is complex-conjugated:

>>> np.dot([2j, 3j], [2j, 3j])
(-13+0j)

For 2-D arrays it’s the matrix product:

>>> a = [[1, 0], [0, 1]]
>>> b = [[4, 1], [2, 2]]
>>> np.dot(a, b)
array([[4, 1],
[2, 2]])

>>> a = np.arange(3*4*5*6).reshape((3,4,5,6))
>>> b = np.arange(3*4*5*6)[::-1].reshape((5,4,6,3))
>>> np.dot(a, b)[2,3,2,1,2,2]
499128

22 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> sum(a[2,3,2,:] * b[1,2,:,2])

499128

numpy.mean(a, axis=None, dtype=None, out=None)

Compute the arithmetic mean along the specified axis.
Returns the average of the array elements. The average is taken over the flattened array by default, other-
wise over the specified axis. float64 intermediate and return values are used for integer inputs.
Parameters
a : array_like
Array containing numbers whose mean is desired. If a is not an array, a conversion is
attempted.
axis : int, optional
Axis along which the means are computed. The default is to compute the mean of the
flattened array.
dtype : data-type, optional
Type to use in computing the mean. For integer inputs, the default is float64; for floating
point inputs, it is the same as the input dtype.
out : ndarray, optional
Alternate output array in which to place the result. The default is None; if provided, it
must have the same shape as the expected output, but the type will be cast if necessary.
See doc.ufuncs for details.
Returns
m : ndarray, see dtype parameter above
If out=None, returns a new array containing the mean values, otherwise a reference to
the output array is returned.
See Also:

average
Weighted average

Notes
The arithmetic mean is the sum of the elements along the axis divided by the number of elements.
Note that for floating-point input, the mean is computed using the same precision the input has. Depending
on the input data, this can cause the results to be inaccurate, especially for float32 (see example below).
Specifying a higher-precision accumulator using the dtype keyword can alleviate this issue.

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.mean(a)
2.5
>>> np.mean(a, axis=0)
array([ 2., 3.])
>>> np.mean(a, axis=1)
array([ 1.5, 3.5])

In single precision, mean can be inaccurate:

1.1. The N-dimensional array (ndarray) 23

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a = np.zeros((2, 512*512), dtype=np.float32)

>>> a[0, :] = 1.0
>>> a[1, :] = 0.1
>>> np.mean(a)
0.546875

Computing the mean in float64 is more accurate:

>>> np.mean(a, dtype=np.float64)
0.55000000074505806

numpy.nonzero(a)
Return the indices of the elements that are non-zero.
Returns a tuple of arrays, one for each dimension of a, containing the indices of the non-zero elements in
that dimension. The corresponding non-zero values can be obtained with:
a[nonzero(a)]

To group the indices by element, rather than dimension, use:

transpose(nonzero(a))

The result of this is always a 2-D array, with a row for each non-zero element.
Parameters
a : array_like
Input array.
Returns
tuple_of_arrays : tuple
Indices of elements that are non-zero.
See Also:

flatnonzero
Return indices that are non-zero in the flattened version of the input array.
ndarray.nonzero
Equivalent ndarray method.
count_nonzero
Counts the number of non-zero elements in the input array.

Examples
>>> x = np.eye(3)
>>> x
array([[ 1., 0., 0.],
[ 0., 1., 0.],
[ 0., 0., 1.]])
>>> np.nonzero(x)
(array([0, 1, 2]), array([0, 1, 2]))

>>> x[np.nonzero(x)]
array([ 1., 1., 1.])
>>> np.transpose(np.nonzero(x))
array([[0, 0],
[1, 1],
[2, 2]])

24 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

A common use for nonzero is to find the indices of an array, where a condition is True. Given an array
a, the condition a > 3 is a boolean array and since False is interpreted as 0, np.nonzero(a > 3) yields the
indices of the a where the condition is true.
>>> a = np.array([[1,2,3],[4,5,6],[7,8,9]])
>>> a > 3
array([[False, False, False],
[ True, True, True],
[ True, True, True]], dtype=bool)
>>> np.nonzero(a > 3)
(array([1, 1, 1, 2, 2, 2]), array([0, 1, 2, 0, 1, 2]))

The nonzero method of the boolean array can also be called.

>>> (a > 3).nonzero()
(array([1, 1, 1, 2, 2, 2]), array([0, 1, 2, 0, 1, 2]))

numpy.prod(a, axis=None, dtype=None, out=None)

Return the product of array elements over a given axis.
Parameters
a : array_like
Input data.
axis : int, optional
Axis over which the product is taken. By default, the product of all elements is calcu-
lated.
dtype : data-type, optional
The data-type of the returned array, as well as of the accumulator in which the elements
are multiplied. By default, if a is of integer type, dtype is the default platform integer.
(Note: if the type of a is unsigned, then so is dtype.) Otherwise, the dtype is the same
as that of a.
out : ndarray, optional
Alternative output array in which to place the result. It must have the same shape as the
expected output, but the type of the output values will be cast if necessary.
Returns
product_along_axis : ndarray, see dtype parameter above.
An array shaped as a but with the specified axis removed. Returns a reference to out if
specified.
See Also:

ndarray.prod
equivalent method
numpy.doc.ufuncs
Section “Output arguments”

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow. That means that, on a
32-bit platform:

1.1. The N-dimensional array (ndarray) 25

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x = np.array([536870910, 536870910, 536870910, 536870910])

>>> np.prod(x) #random
16

Examples
By default, calculate the product of all elements:
>>> np.prod([1.,2.])
2.0

Even when the input array is two-dimensional:

>>> np.prod([[1.,2.],[3.,4.]])
24.0

But we can also specify the axis over which to multiply:

>>> np.prod([[1.,2.],[3.,4.]], axis=1)
array([ 2., 12.])

If the type of x is unsigned, then the output type is the unsigned platform integer:
>>> x = np.array([1, 2, 3], dtype=np.uint8)
>>> np.prod(x).dtype == np.uint
True

If x is of a signed integer type, then the output type is the default platform integer:
>>> x = np.array([1, 2, 3], dtype=np.int8)
>>> np.prod(x).dtype == np.int
True

numpy.ptp(a, axis=None, out=None)

Range of values (maximum - minimum) along an axis.
The name of the function comes from the acronym for ‘peak to peak’.
Parameters
a : array_like
Input values.
axis : int, optional
Axis along which to find the peaks. By default, flatten the array.
out : array_like
Alternative output array in which to place the result. It must have the same shape and
buffer length as the expected output, but the type of the output values will be cast if
necessary.
Returns
ptp : ndarray
A new array holding the result, unless out was specified, in which case a reference to
out is returned.

Examples

26 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.ptp(x, axis=0)

array([2, 2])

>>> np.ptp(x, axis=1)

array([1, 1])

numpy.put(a, ind, v, mode=’raise’)

Replaces specified elements of an array with given values.
The indexing works on the flattened target array. put is roughly equivalent to:
a.flat[ind] = v

Parameters
a : ndarray
Target array.
ind : array_like
Target indices, interpreted as integers.
v : array_like
Values to place in a at target indices. If v is shorter than ind it will be repeated as
necessary.
mode : {‘raise’, ‘wrap’, ‘clip’}, optional
Specifies how out-of-bounds indices will behave.
• ‘raise’ – raise an error (default)
• ‘wrap’ – wrap around
• ‘clip’ – clip to the range
‘clip’ mode means that all indices that are too large are replaced by the index that ad-
dresses the last element along that axis. Note that this disables indexing with negative
numbers.

See Also:
putmask, place

Examples
>>> a = np.arange(5)
>>> np.put(a, [0, 2], [-44, -55])
>>> a
array([-44, 1, -55, 3, 4])

>>> a = np.arange(5)
>>> np.put(a, 22, -5, mode=’clip’)
>>> a
array([ 0, 1, 2, 3, -5])

1.1. The N-dimensional array (ndarray) 27

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.ravel(a, order=’C’)
Return a flattened array.
A 1-D array, containing the elements of the input, is returned. A copy is made only if needed.
Parameters
a : array_like
Input array. The elements in a are read in the order specified by order, and packed as a
1-D array.
order : {‘C’,’F’, ‘A’, ‘K’}, optional
The elements of a are read in this order. ‘C’ means to view the elements in C (row-
major) order. ‘F’ means to view the elements in Fortran (column-major) order. ‘A’
means to view the elements in ‘F’ order if a is Fortran contiguous, ‘C’ order otherwise.
‘K’ means to view the elements in the order they occur in memory, except for reversing
the data when strides are negative. By default, ‘C’ order is used.
Returns
1d_array : ndarray
Output of the same dtype as a, and of shape (a.size(),).
See Also:

ndarray.flat
1-D iterator over an array.
ndarray.flatten
1-D array copy of the elements of an array in row-major order.

Notes
In row-major order, the row index varies the slowest, and the column index the quickest. This can be
generalized to multiple dimensions, where row-major order implies that the index along the first axis
varies slowest, and the index along the last quickest. The opposite holds for Fortran-, or column-major,
mode.

Examples
It is equivalent to reshape(-1, order=order).
>>> x = np.array([[1, 2, 3], [4, 5, 6]])
>>> print np.ravel(x)
[1 2 3 4 5 6]

>>> print x.reshape(-1)

[1 2 3 4 5 6]

>>> print np.ravel(x, order=’F’)

[1 4 2 5 3 6]

When order is ‘A’, it will preserve the array’s ‘C’ or ‘F’ ordering:
>>> print np.ravel(x.T)
[1 4 2 5 3 6]
>>> print np.ravel(x.T, order=’A’)
[1 2 3 4 5 6]

When order is ‘K’, it will preserve orderings that are neither ‘C’ nor ‘F’, but won’t reverse axes:

28 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a = np.arange(3)[::-1]; a
array([2, 1, 0])
>>> a.ravel(order=’C’)
array([2, 1, 0])
>>> a.ravel(order=’K’)
array([2, 1, 0])

>>> a = np.arange(12).reshape(2,3,2).swapaxes(1,2); a
array([[[ 0, 2, 4],
[ 1, 3, 5]],
[[ 6, 8, 10],
[ 7, 9, 11]]])
>>> a.ravel(order=’C’)
array([ 0, 2, 4, 1, 3, 5, 6, 8, 10, 7, 9, 11])
>>> a.ravel(order=’K’)
array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11])

numpy.repeat(a, repeats, axis=None)

Repeat elements of an array.
Parameters
a : array_like
Input array.
repeats : {int, array of ints}
The number of repetitions for each element. repeats is broadcasted to fit the shape of
the given axis.
axis : int, optional
The axis along which to repeat values. By default, use the flattened input array, and
return a flat output array.
Returns
repeated_array : ndarray
Output array which has the same shape as a, except along the given axis.
See Also:

tile
Tile an array.

Examples
>>> x = np.array([[1,2],[3,4]])
>>> np.repeat(x, 2)
array([1, 1, 2, 2, 3, 3, 4, 4])
>>> np.repeat(x, 3, axis=1)
array([[1, 1, 1, 2, 2, 2],
[3, 3, 3, 4, 4, 4]])
>>> np.repeat(x, [1, 2], axis=0)
array([[1, 2],
[3, 4],
[3, 4]])

numpy.reshape(a, newshape, order=’C’)

Gives a new shape to an array without changing its data.

1.1. The N-dimensional array (ndarray) 29

NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
a : array_like
Array to be reshaped.
newshape : int or tuple of ints
The new shape should be compatible with the original shape. If an integer, then the
result will be a 1-D array of that length. One shape dimension can be -1. In this case,
the value is inferred from the length of the array and remaining dimensions.
order : {‘C’, ‘F’, ‘A’}, optional
Determines whether the array data should be viewed as in C (row-major) order, FOR-
TRAN (column-major) order, or the C/FORTRAN order should be preserved.
Returns
reshaped_array : ndarray
This will be a new view object if possible; otherwise, it will be a copy.
See Also:

ndarray.reshape
Equivalent method.

Notes
It is not always possible to change the shape of an array without copying the data. If you want an error to
be raise if the data is copied, you should assign the new shape to the shape attribute of the array:
>>> a = np.zeros((10, 2))
# A transpose make the array non-contiguous
>>> b = a.T
# Taking a view makes it possible to modify the shape without modiying the
# initial object.
>>> c = b.view()
>>> c.shape = (20)
AttributeError: incompatible shape for a non-contiguous array

Examples
>>> a = np.array([[1,2,3], [4,5,6]])
>>> np.reshape(a, 6)
array([1, 2, 3, 4, 5, 6])
>>> np.reshape(a, 6, order=’F’)
array([1, 4, 2, 5, 3, 6])

>>> np.reshape(a, (3,-1)) # the unspecified value is inferred to be 2

array([[1, 2],
[3, 4],
[5, 6]])

30 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Array to be resized.
new_shape : int or tuple of int
Shape of resized array.
Returns
reshaped_array : ndarray
The new array is formed from the data in the old array, repeated if necessary to fill out
the required number of elements. The data are repeated in the order that they are stored
in memory.
See Also:

ndarray.resize
resize an array in-place.

Examples
>>> a=np.array([[0,1],[2,3]])
>>> np.resize(a,(1,4))
array([[0, 1, 2, 3]])
>>> np.resize(a,(2,4))
array([[0, 1, 2, 3],
[0, 1, 2, 3]])

numpy.searchsorted(a, v, side=’left’)
Find indices where elements should be inserted to maintain order.
Find the indices into a sorted array a such that, if the corresponding elements in v were inserted before the
indices, the order of a would be preserved.
Parameters
a : 1-D array_like
Input array, sorted in ascending order.
v : array_like
Values to insert into a.
side : {‘left’, ‘right’}, optional
If ‘left’, the index of the first suitable location found is given. If ‘right’, return the last
such index. If there is no suitable index, return either 0 or N (where N is the length of
a).
Returns
indices : array of ints
Array of insertion points with the same shape as v.
See Also:

sort
Return a sorted copy of an array.
histogram
Produce histogram from 1-D data.

1.1. The N-dimensional array (ndarray) 31

NumPy Reference, Release 2.0.0.dev-ac3cba3

Notes
Binary search is used to find the required insertion points.
As of Numpy 1.4.0 searchsorted works with real/complex arrays containing nan values. The enhanced
sort order is documented in sort.

Examples
>>> np.searchsorted([1,2,3,4,5], 3)
2
>>> np.searchsorted([1,2,3,4,5], 3, side=’right’)
3
>>> np.searchsorted([1,2,3,4,5], [-10, 10, 2, 3])
array([0, 5, 1, 2])

numpy.sort(a, axis=-1, kind=’quicksort’, order=None)

Return a sorted copy of an array.
Parameters
a : array_like
Array to be sorted.
axis : int or None, optional
Axis along which to sort. If None, the array is flattened before sorting. The default is
-1, which sorts along the last axis.
kind : {‘quicksort’, ‘mergesort’, ‘heapsort’}, optional
Sorting algorithm. Default is ‘quicksort’.
order : list, optional
When a is a structured array, this argument specifies which fields to compare first, sec-
ond, and so on. This list does not need to include all of the fields.
Returns
sorted_array : ndarray
Array of the same type and shape as a.
See Also:

ndarray.sort
Method to sort an array in-place.
argsort
Indirect sort.
lexsort
Indirect stable sort on multiple keys.
searchsorted
Find elements in a sorted array.

32 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

kind speed worst case work space stable

‘quicksort’ 1 O(n^2) 0 no
‘mergesort’ 2 O(n*log(n)) ~n/2 yes
‘heapsort’ 3 O(n*log(n)) 0 no
All the sort algorithms make temporary copies of the data when sorting along any but the last axis. Con-
sequently, sorting along the last axis is faster and uses less space than sorting along any other axis.
The sort order for complex numbers is lexicographic. If both the real and imaginary parts are non-nan then
the order is determined by the real parts except when they are equal, in which case the order is determined
by the imaginary parts.
Previous to numpy 1.4.0 sorting real and complex arrays containing nan values led to undefined behaviour.
In numpy versions >= 1.4.0 nan values are sorted to the end. The extended sort order is:
•Real: [R, nan]
•Complex: [R + Rj, R + nanj, nan + Rj, nan + nanj]
where R is a non-nan real value. Complex values with the same nan placements are sorted according to
the non-nan part if it exists. Non-nan values are sorted as before.

Examples
>>> a = np.array([[1,4],[3,1]])
>>> np.sort(a) # sort along the last axis
array([[1, 4],
[1, 3]])
>>> np.sort(a, axis=None) # sort the flattened array
array([1, 1, 3, 4])
>>> np.sort(a, axis=0) # sort along the first axis
array([[1, 1],
[3, 4]])

Use the order keyword to specify a field to use when sorting a structured array:
>>> dtype = [(’name’, ’S10’), (’height’, float), (’age’, int)]
>>> values = [(’Arthur’, 1.8, 41), (’Lancelot’, 1.9, 38),
... (’Galahad’, 1.7, 38)]
>>> a = np.array(values, dtype=dtype) # create a structured array
>>> np.sort(a, order=’height’)
array([(’Galahad’, 1.7, 38), (’Arthur’, 1.8, 41),
(’Lancelot’, 1.8999999999999999, 38)],
dtype=[(’name’, ’|S10’), (’height’, ’<f8’), (’age’, ’<i4’)])

Sort by age, then height if ages are equal:

>>> np.sort(a, order=[’age’, ’height’])
array([(’Galahad’, 1.7, 38), (’Lancelot’, 1.8999999999999999, 38),
(’Arthur’, 1.8, 41)],
dtype=[(’name’, ’|S10’), (’height’, ’<f8’), (’age’, ’<i4’)])

numpy.squeeze(a)
Remove single-dimensional entries from the shape of an array.
Parameters
a : array_like
Input data.
Returns
squeezed : ndarray

1.1. The N-dimensional array (ndarray) 33

NumPy Reference, Release 2.0.0.dev-ac3cba3

The input array, but with with all dimensions of length 1 removed. Whenever possible,
a view on a is returned.

Examples
>>> x = np.array([[[0], [1], [2]]])
>>> x.shape
(1, 3, 1)
>>> np.squeeze(x).shape
(3,)

numpy.std(a, axis=None, dtype=None, out=None, ddof=0)

Compute the standard deviation along the specified axis.
Returns the standard deviation, a measure of the spread of a distribution, of the array elements. The
standard deviation is computed for the flattened array by default, otherwise over the specified axis.
Parameters
a : array_like
Calculate the standard deviation of these values.
axis : int, optional
Axis along which the standard deviation is computed. The default is to compute the
standard deviation of the flattened array.
dtype : dtype, optional
Type to use in computing the standard deviation. For arrays of integer type the default
is float64, for arrays of float types it is the same as the array type.
out : ndarray, optional
Alternative output array in which to place the result. It must have the same shape as the
expected output but the type (of the calculated values) will be cast if necessary.
ddof : int, optional
Means Delta Degrees of Freedom. The divisor used in calculations is N - ddof,
where N represents the number of elements. By default ddof is zero.
Returns
standard_deviation : ndarray, see dtype parameter above.
If out is None, return a new array containing the standard deviation, otherwise return a
reference to the output array.
See Also:
var, mean
numpy.doc.ufuncs
Section “Output arguments”

34 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

this function is the square root of the estimated variance, so even with ddof=1, it will not be an unbiased
estimate of the standard deviation per se.
Note that, for complex numbers, std takes the absolute value before squaring, so that the result is always
real and nonnegative.
For floating-point input, the std is computed using the same precision the input has. Depending on the input
data, this can cause the results to be inaccurate, especially for float32 (see example below). Specifying a
higher-accuracy accumulator using the dtype keyword can alleviate this issue.

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.std(a)
1.1180339887498949
>>> np.std(a, axis=0)
array([ 1., 1.])
>>> np.std(a, axis=1)
array([ 0.5, 0.5])

In single precision, std() can be inaccurate:

>>> a = np.zeros((2,512*512), dtype=np.float32)
>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.std(a)
0.45172946707416706

Computing the standard deviation in float64 is more accurate:

>>> np.std(a, dtype=np.float64)
0.44999999925552653

numpy.sum(a, axis=None, dtype=None, out=None)

Sum of array elements over a given axis.
Parameters
a : array_like
Elements to sum.
axis : integer, optional
Axis over which the sum is taken. By default axis is None, and all elements are summed.
dtype : dtype, optional
The type of the returned array and of the accumulator in which the elements are
summed. By default, the dtype of a is used. An exception is when a has an integer
type with less precision than the default platform integer. In that case, the default plat-
form integer is used instead.
out : ndarray, optional
Array into which the output is placed. By default, a new array is created. If out is
given, it must be of the appropriate shape (the shape of a with axis removed, i.e.,
numpy.delete(a.shape, axis)). Its type is preserved. See doc.ufuncs (Sec-
tion “Output arguments”) for more details.
Returns
sum_along_axis : ndarray

1.1. The N-dimensional array (ndarray) 35

NumPy Reference, Release 2.0.0.dev-ac3cba3

An array with the same shape as a, with the specified axis removed. If a is a 0-d array,
or if axis is None, a scalar is returned. If an output array is specified, a reference to out
is returned.
See Also:

ndarray.sum
Equivalent method.
cumsum
Cumulative sum of array elements.
trapz
Integration of array values using the composite trapezoidal rule.

mean, average

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

Examples
>>> np.sum([0.5, 1.5])
2.0
>>> np.sum([0.5, 0.7, 0.2, 1.5], dtype=np.int32)
1
>>> np.sum([[0, 1], [0, 5]])
6
>>> np.sum([[0, 1], [0, 5]], axis=0)
array([0, 6])
>>> np.sum([[0, 1], [0, 5]], axis=1)
array([1, 5])

If the accumulator is too small, overflow occurs:

>>> np.ones(128, dtype=np.int8).sum(dtype=np.int8)
-128

numpy.swapaxes(a, axis1, axis2)

Interchange two axes of an array.
Parameters
a : array_like
Input array.
axis1 : int
First axis.
axis2 : int
Second axis.
Returns
a_swapped : ndarray
If a is an ndarray, then a view of a is returned; otherwise a new array is created.

36 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> x = np.array([[1,2,3]])
>>> np.swapaxes(x,0,1)
array([[1],
[2],
[3]])

>>> x = np.array([[[0,1],[2,3]],[[4,5],[6,7]]])
>>> x
array([[[0, 1],
[2, 3]],
[[4, 5],
[6, 7]]])

>>> np.swapaxes(x,0,2)
array([[[0, 4],
[2, 6]],
[[1, 5],
[3, 7]]])

numpy.take(a, indices, axis=None, out=None, mode=’raise’)

Take elements from an array along an axis.
This function does the same thing as “fancy” indexing (indexing arrays using arrays); however, it can be
easier to use if you need elements along a given axis.
Parameters
a : array_like
The source array.
indices : array_like
The indices of the values to extract.
axis : int, optional
The axis over which to select values. By default, the flattened input array is used.
out : ndarray, optional
If provided, the result will be placed in this array. It should be of the appropriate shape
and dtype.
mode : {‘raise’, ‘wrap’, ‘clip’}, optional
Specifies how out-of-bounds indices will behave.
• ‘raise’ – raise an error (default)
• ‘wrap’ – wrap around
• ‘clip’ – clip to the range
‘clip’ mode means that all indices that are too large are replaced by the index that ad-
dresses the last element along that axis. Note that this disables indexing with negative
numbers.
Returns
subarray : ndarray
The returned array has the same type as a.
See Also:

1.1. The N-dimensional array (ndarray) 37

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.take
equivalent method

Examples
>>> a = [4, 3, 5, 7, 6, 8]
>>> indices = [0, 1, 4]
>>> np.take(a, indices)
array([4, 3, 6])

In this example if a is an ndarray, “fancy” indexing can be used.

>>> a = np.array(a)
>>> a[indices]
array([4, 3, 6])

numpy.trace(a, offset=0, axis1=0, axis2=1, dtype=None, out=None)

Return the sum along diagonals of the array.
If a is 2-D, the sum along its diagonal with the given offset is returned, i.e., the sum of elements
a[i,i+offset] for all i.
If a has more than two dimensions, then the axes specified by axis1 and axis2 are used to determine the
2-D sub-arrays whose traces are returned. The shape of the resulting array is the same as that of a with
axis1 and axis2 removed.
Parameters
a : array_like
Input array, from which the diagonals are taken.
offset : int, optional
Offset of the diagonal from the main diagonal. Can be both positive and negative.
Defaults to 0.
axis1, axis2 : int, optional
Axes to be used as the first and second axis of the 2-D sub-arrays from which the diag-
onals should be taken. Defaults are the first two axes of a.
dtype : dtype, optional
Determines the data-type of the returned array and of the accumulator where the ele-
ments are summed. If dtype has the value None and a is of integer type of precision less
than the default integer precision, then the default integer precision is used. Otherwise,
the precision is the same as that of a.
out : ndarray, optional
Array into which the output is placed. Its type is preserved and it must be of the right
shape to hold the output.
Returns
sum_along_diagonals : ndarray
If a is 2-D, the sum along the diagonal is returned. If a has larger dimensions, then an
array of sums along diagonals is returned.
See Also:
diag, diagonal, diagflat

38 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> np.trace(np.eye(3))
3.0
>>> a = np.arange(8).reshape((2,2,2))
>>> np.trace(a)
array([6, 8])

>>> a = np.arange(24).reshape((2,2,2,3))
>>> np.trace(a).shape
(2, 3)

numpy.transpose(a, axes=None)
Permute the dimensions of an array.
Parameters
a : array_like
Input array.
axes : list of ints, optional
By default, reverse the dimensions, otherwise permute the axes according to the values
given.
Returns
p : ndarray
a with its axes permuted. A view is returned whenever possible.
See Also:
rollaxis

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.transpose(x)
array([[0, 2],
[1, 3]])

>>> x = np.ones((1, 2, 3))

>>> np.transpose(x, (1, 0, 2)).shape
(2, 1, 3)

numpy.var(a, axis=None, dtype=None, out=None, ddof=0)

1.1. The N-dimensional array (ndarray) 39

NumPy Reference, Release 2.0.0.dev-ac3cba3

Axis along which the variance is computed. The default is to compute the variance of
the flattened array.
dtype : data-type, optional
Type to use in computing the variance. For arrays of integer type the default is float32;
for arrays of float types it is the same as the array type.
out : ndarray, optional
Alternate output array in which to place the result. It must have the same shape as the
expected output, but the type is cast if necessary.
ddof : int, optional
“Delta Degrees of Freedom”: the divisor used in the calculation is N - ddof, where
N represents the number of elements. By default ddof is zero.
Returns
variance : ndarray, see dtype parameter above
If out=None, returns a new array containing the variance; otherwise, a reference to
the output array is returned.
See Also:

std
Standard deviation
mean
Average
numpy.doc.ufuncs
Section “Output arguments”

Notes
The variance is the average of the squared deviations from the mean, i.e., var = mean(abs(x -
x.mean())**2).
The mean is normally calculated as x.sum() / N, where N = len(x). If, however, ddof is specified,
the divisor N - ddof is used instead. In standard statistical practice, ddof=1 provides an unbiased
estimator of the variance of a hypothetical infinite population. ddof=0 provides a maximum likelihood
estimate of the variance for normally distributed variables.
Note that for complex numbers, the absolute value is taken before squaring, so that the result is always real
and nonnegative.
For floating-point input, the variance is computed using the same precision the input has. Depending
on the input data, this can cause the results to be inaccurate, especially for float32 (see example below).
Specifying a higher-accuracy accumulator using the dtype keyword can alleviate this issue.

Examples
>>> a = np.array([[1,2],[3,4]])
>>> np.var(a)
1.25
>>> np.var(a,0)
array([ 1., 1.])
>>> np.var(a,1)
array([ 0.25, 0.25])

40 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

In single precision, var() can be inaccurate:

>>> a = np.zeros((2,512*512), dtype=np.float32)
>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.var(a)
0.20405951142311096

Computing the standard deviation in float64 is more accurate:

>>> np.var(a, dtype=np.float64)
0.20249999932997387
>>> ((1-0.55)**2 + (0.1-0.55)**2)/2
0.20250000000000001

1.1.2 Indexing arrays

Arrays can be indexed using an extended Python slicing syntax, array[selection]. Similar syntax is also used
for accessing fields in a record array.
See Also:
Array Indexing.

1.1.3 Internal memory layout of an ndarray

An instance of class ndarray consists of a contiguous one-dimensional segment of computer memory (owned by
the array, or by some other object), combined with an indexing scheme that maps N integers into the location of an
item in the block. The ranges in which the indices can vary is specified by the shape of the array. How many
bytes each item takes and how the bytes are interpreted is defined by the data-type object associated with the array.
A segment of memory is inherently 1-dimensional, and there are many different schemes for arranging the items of
an N-dimensional array in a 1-dimensional block. Numpy is flexible, and ndarray objects can accommodate any
strided indexing scheme. In a strided scheme, the N-dimensional index (n0 , n1 , ..., nN −1 ) corresponds to the offset
(in bytes):
N
X −1
noffset = sk nk
k=0

from the beginning of the memory block associated with the array. Here, sk are integers which specify the strides
of the array. The column-major order (used, for example, in the Fortran language and in Matlab) and row-major order
(used in C) schemes are just specific kinds of strided scheme, and correspond to the strides:
k−1
Y N
Y −1
scolumn
k = dj , srow
k = dj .
j=0 j=k+1

where dj = self.itemsize * self.shape[j].

Both the C and Fortran orders are contiguous, i.e., single-segment, memory layouts, in which every part of the memory
block can be accessed by some combination of the indices.
Data in new ndarrays is in the row-major (C) order, unless otherwise specified, but, for example, basic array slicing
often produces views in a different scheme.
Note: Several algorithms in NumPy work on arbitrarily strided arrays. However, some algorithms require single-
segment arrays. When an irregularly strided array is passed in to such algorithms, a copy is automatically made.

1.1. The N-dimensional array (ndarray) 41

NumPy Reference, Release 2.0.0.dev-ac3cba3

1.1.4 Array attributes

Array attributes reflect information that is intrinsic to the array itself. Generally, accessing an array through its at-
tributes allows you to get and sometimes set intrinsic properties of the array without creating a new array. The exposed
attributes are the core parts of an array and only some of them can be reset meaningfully without creating a new array.
Information on each attribute is given below.

Memory layout

The following attributes contain information about the memory layout of the array:
ndarray.flags Information about the memory layout of the array.
ndarray.shape Tuple of array dimensions.
ndarray.strides Tuple of bytes to step in each dimension when traversing an array.
ndarray.ndim Number of array dimensions.
ndarray.data Python buffer object pointing to the start of the array’s data.
ndarray.size Number of elements in the array.
ndarray.itemsize Length of one array element in bytes.
ndarray.nbytes Total bytes consumed by the elements of the array.
ndarray.base Base object if memory is from some other object.
ndarray.flags
Information about the memory layout of the array.

Notes
The flags object can be accessed dictionary-like (as in a.flags[’WRITEABLE’]), or by using lowercased
attribute names (as in a.flags.writeable). Short flag names are only supported in dictionary access.
Only the UPDATEIFCOPY, WRITEABLE, and ALIGNED flags can be changed by the user, via direct assign-
ment to the attribute or dictionary entry, or by calling ndarray.setflags.
The array flags cannot be set arbitrarily:
•UPDATEIFCOPY can only be set False.
•ALIGNED can only be set True if the data is truly aligned.
•WRITEABLE can only be set True if the array owns its own memory or the ultimate owner of the memory
exposes a writeable buffer interface or is a string.

Attributes
ndarray.shape
Tuple of array dimensions.

Notes
May be used to “reshape” the array, as long as this would not require a change in the total number of elements

Examples
>>> x = np.array([1, 2, 3, 4])
>>> x.shape
(4,)
>>> y = np.zeros((2, 3, 4))
>>> y.shape
(2, 3, 4)
>>> y.shape = (3, 8)
>>> y

42 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

array([[ 0., 0., 0., 0., 0., 0., 0., 0.],

[ 0., 0., 0., 0., 0., 0., 0., 0.],
[ 0., 0., 0., 0., 0., 0., 0., 0.]])
>>> y.shape = (3, 6)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ValueError: total size of new array must be unchanged

ndarray.strides
Tuple of bytes to step in each dimension when traversing an array.
The byte offset of element (i[0], i[1], ..., i[n]) in an array a is:
offset = sum(np.array(i) * a.strides)

A more detailed explanation of strides can be found in the “ndarray.rst” file in the NumPy reference guide.
See Also:
numpy.lib.stride_tricks.as_strided

Notes
Imagine an array of 32-bit integers (each 4 bytes):
x = np.array([[0, 1, 2, 3, 4],
[5, 6, 7, 8, 9]], dtype=np.int32)

This array is stored in memory as 40 bytes, one after the other (known as a contiguous block of memory). The
strides of an array tell us how many bytes we have to skip in memory to move to the next position along a certain
axis. For example, we have to skip 4 bytes (1 value) to move to the next column, but 20 bytes (5 values) to get
to the same position in the next row. As such, the strides for the array x will be (20, 4).

Examples
>>> y = np.reshape(np.arange(2*3*4), (2,3,4))
>>> y
array([[[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11]],
[[12, 13, 14, 15],
[16, 17, 18, 19],
[20, 21, 22, 23]]])
>>> y.strides
(48, 16, 4)
>>> y[1,1,1]
17
>>> offset=sum(y.strides * np.array((1,1,1)))
>>> offset/y.itemsize
17

>>> x = np.reshape(np.arange(567*8), (5,6,7,8)).transpose(2,3,1,0)

>>> x.strides
(32, 4, 224, 1344)
>>> i = np.array([3,5,2,2])
>>> offset = sum(i * x.strides)
>>> x[3,5,2,2]
813
>>> offset / x.itemsize
813

1.1. The N-dimensional array (ndarray) 43

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.ndim
Number of array dimensions.

Examples
>>> x = np.array([1, 2, 3])
>>> x.ndim
1
>>> y = np.zeros((2, 3, 4))
>>> y.ndim
3

ndarray.data
Python buffer object pointing to the start of the array’s data.
ndarray.size
Number of elements in the array.
Equivalent to np.prod(a.shape), i.e., the product of the array’s dimensions.

Examples
>>> x = np.zeros((3, 5, 2), dtype=np.complex128)
>>> x.size
30
>>> np.prod(x.shape)
30

ndarray.itemsize
Length of one array element in bytes.

Examples
>>> x = np.array([1,2,3], dtype=np.float64)
>>> x.itemsize
8
>>> x = np.array([1,2,3], dtype=np.complex128)
>>> x.itemsize
16

ndarray.nbytes
Total bytes consumed by the elements of the array.

Notes
Does not include memory consumed by non-element attributes of the array object.

Examples
>>> x = np.zeros((3,5,2), dtype=np.complex128)
>>> x.nbytes
480
>>> np.prod(x.shape) * x.itemsize
480

ndarray.base
Base object if memory is from some other object.

Examples
The base of an array that owns its memory is None:

44 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x = np.array([1,2,3,4])
>>> x.base is None
True

Slicing creates a view, whose memory is shared with x:

>>> y = x[2:]
>>> y.base is x
True

Data type

See Also:
Data type objects
The data type object associated with the array can be found in the dtype attribute:
ndarray.dtype Data-type of the array’s elements.
ndarray.dtype
Data-type of the array’s elements.
Parameters
None :
Returns
d : numpy dtype object
See Also:
numpy.dtype

Examples
>>> x
array([[0, 1],
[2, 3]])
>>> x.dtype
dtype(’int32’)
>>> type(x.dtype)
<type ’numpy.dtype’>

Other attributes

ndarray.T Same as self.transpose(), except that self is returned if self.ndim < 2.

ndarray.real The real part of the array.
ndarray.imag The imaginary part of the array.
ndarray.flat A 1-D iterator over the array.
ndarray.ctypes An object to simplify the interaction of the array with the ctypes module.
__array_priority__
ndarray.T
Same as self.transpose(), except that self is returned if self.ndim < 2.

Examples

1.1. The N-dimensional array (ndarray) 45

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x = np.array([[1.,2.],[3.,4.]])
>>> x
array([[ 1., 2.],
[ 3., 4.]])
>>> x.T
array([[ 1., 3.],
[ 2., 4.]])
>>> x = np.array([1.,2.,3.,4.])
>>> x
array([ 1., 2., 3., 4.])
>>> x.T
array([ 1., 2., 3., 4.])

ndarray.real
The real part of the array.
See Also:

numpy.real
equivalent function

Examples
>>> x = np.sqrt([1+0j, 0+1j])
>>> x.real
array([ 1. , 0.70710678])
>>> x.real.dtype
dtype(’float64’)

ndarray.imag
The imaginary part of the array.

Examples
>>> x = np.sqrt([1+0j, 0+1j])
>>> x.imag
array([ 0. , 0.70710678])
>>> x.imag.dtype
dtype(’float64’)

ndarray.flat
A 1-D iterator over the array.
This is a numpy.flatiter instance, which acts similarly to, but is not a subclass of, Python’s built-in iterator
object.
See Also:

flatten
Return a copy of the array collapsed into one dimension.

flatiter

Examples
>>> x = np.arange(1, 7).reshape(2, 3)
>>> x
array([[1, 2, 3],
[4, 5, 6]])

46 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x.flat[3]
4
>>> x.T
array([[1, 4],
[2, 5],
[3, 6]])
>>> x.T.flat[3]
5
>>> type(x.flat)
<type ’numpy.flatiter’>

An assignment example:
>>> x.flat = 3; x
array([[3, 3, 3],
[3, 3, 3]])
>>> x.flat[[1,4]] = 1; x
array([[3, 1, 3],
[3, 1, 3]])

ndarray.ctypes
An object to simplify the interaction of the array with the ctypes module.
This attribute creates an object that makes it easier to use arrays when calling shared libraries with the ctypes
module. The returned object has, among others, data, shape, and strides attributes (see Notes below) which
themselves return ctypes objects that can be used as arguments to a shared library.
Parameters
None :
Returns
c : Python object
Possessing attributes data, shape, strides, etc.
See Also:
numpy.ctypeslib

1.1. The N-dimensional array (ndarray) 47

NumPy Reference, Release 2.0.0.dev-ac3cba3

•data_as(obj): Return the data pointer cast to a particular c-types object. For example, calling
self._as_parameter_ is equivalent to self.data_as(ctypes.c_void_p). Perhaps you want to use the data as
a pointer to a ctypes array of floating-point data: self.data_as(ctypes.POINTER(ctypes.c_double)).
•shape_as(obj): Return the shape tuple as an array of some other c-types type. For example:
self.shape_as(ctypes.c_short).
•strides_as(obj): Return the strides tuple as an array of some other c-types type. For example:
self.strides_as(ctypes.c_longlong).
Be careful using the ctypes attribute - especially on temporary arrays or arrays constructed on the fly. For
example, calling (a+b).ctypes.data_as(ctypes.c_void_p) returns a pointer to memory that is
invalid because the array created as (a+b) is deallocated before the next Python statement. You can avoid this
problem using either c=a+b or ct=(a+b).ctypes. In the latter case, ct will hold a reference to the array
until ct is deleted or re-assigned.
If the ctypes module is not available, then the ctypes attribute of array objects still returns something useful, but
ctypes objects are not returned and errors may be raised instead. In particular, the object will still have the as
parameter attribute which will return an integer equal to the data attribute.

Examples
>>> import ctypes
>>> x
array([[0, 1],
[2, 3]])
>>> x.ctypes.data
30439712
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_long))
<ctypes.LP_c_long object at 0x01F01300>
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_long)).contents
c_long(0)
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_longlong)).contents
c_longlong(4294967296L)
>>> x.ctypes.shape
<numpy.core._internal.c_long_Array_2 object at 0x01FFD580>
>>> x.ctypes.shape_as(ctypes.c_long)
<numpy.core._internal.c_long_Array_2 object at 0x01FCE620>
>>> x.ctypes.strides
<numpy.core._internal.c_long_Array_2 object at 0x01FCE620>
>>> x.ctypes.strides_as(ctypes.c_longlong)
<numpy.core._internal.c_longlong_Array_2 object at 0x01F01300>

Array interface

See Also:
The Array Interface.
__array_interface__ Python-side of the array interface
__array_struct__ C-side of the array interface

ctypes foreign function interface

ndarray.ctypes An object to simplify the interaction of the array with the ctypes module.
ndarray.ctypes
An object to simplify the interaction of the array with the ctypes module.

48 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

This attribute creates an object that makes it easier to use arrays when calling shared libraries with the ctypes
module. The returned object has, among others, data, shape, and strides attributes (see Notes below) which
themselves return ctypes objects that can be used as arguments to a shared library.
Parameters
None :
Returns
c : Python object
Possessing attributes data, shape, strides, etc.
See Also:
numpy.ctypeslib

Notes
Below are the public attributes of this object which were documented in “Guide to NumPy” (we have omitted
undocumented public attributes, as well as documented private attributes):
•data: A pointer to the memory area of the array as a Python integer. This memory area may contain data
that is not aligned, or not in correct byte-order. The memory area may not even be writeable. The array
flags and data-type of this array should be respected when passing this attribute to arbitrary C-code to avoid
trouble that can include Python crashing. User Beware! The value of this attribute is exactly the same as
self._array_interface_[’data’][0].
•shape (c_intp*self.ndim): A ctypes array of length self.ndim where the basetype is the C-integer corre-
sponding to dtype(‘p’) on this platform. This base-type could be c_int, c_long, or c_longlong depending
on the platform. The c_intp type is defined accordingly in numpy.ctypeslib. The ctypes array contains the
shape of the underlying array.
•strides (c_intp*self.ndim): A ctypes array of length self.ndim where the basetype is the same as for the
shape attribute. This ctypes array contains the strides information from the underlying array. This strides
information is important for showing how many bytes must be jumped to get to the next element in the
array.
•data_as(obj): Return the data pointer cast to a particular c-types object. For example, calling
self._as_parameter_ is equivalent to self.data_as(ctypes.c_void_p). Perhaps you want to use the data as
a pointer to a ctypes array of floating-point data: self.data_as(ctypes.POINTER(ctypes.c_double)).
•shape_as(obj): Return the shape tuple as an array of some other c-types type. For example:
self.shape_as(ctypes.c_short).
•strides_as(obj): Return the strides tuple as an array of some other c-types type. For example:
self.strides_as(ctypes.c_longlong).
Be careful using the ctypes attribute - especially on temporary arrays or arrays constructed on the fly. For
example, calling (a+b).ctypes.data_as(ctypes.c_void_p) returns a pointer to memory that is
invalid because the array created as (a+b) is deallocated before the next Python statement. You can avoid this
problem using either c=a+b or ct=(a+b).ctypes. In the latter case, ct will hold a reference to the array
until ct is deleted or re-assigned.
If the ctypes module is not available, then the ctypes attribute of array objects still returns something useful, but
ctypes objects are not returned and errors may be raised instead. In particular, the object will still have the as
parameter attribute which will return an integer equal to the data attribute.

Examples
>>> import ctypes
>>> x

1.1. The N-dimensional array (ndarray) 49

NumPy Reference, Release 2.0.0.dev-ac3cba3

array([[0, 1],
[2, 3]])
>>> x.ctypes.data
30439712
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_long))
<ctypes.LP_c_long object at 0x01F01300>
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_long)).contents
c_long(0)
>>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_longlong)).contents
c_longlong(4294967296L)
>>> x.ctypes.shape
<numpy.core._internal.c_long_Array_2 object at 0x01FFD580>
>>> x.ctypes.shape_as(ctypes.c_long)
<numpy.core._internal.c_long_Array_2 object at 0x01FCE620>
>>> x.ctypes.strides
<numpy.core._internal.c_long_Array_2 object at 0x01FCE620>
>>> x.ctypes.strides_as(ctypes.c_longlong)
<numpy.core._internal.c_longlong_Array_2 object at 0x01F01300>

1.1.5 Array methods

An ndarray object has many methods which operate on or with the array in some fashion, typically returning an
array result. These methods are briefly explained below. (Each method’s docstring has a more complete description.)
For the following methods there are also corresponding functions in numpy: all, any, argmax, argmin,
argsort, choose, clip, compress, copy, cumprod, cumsum, diagonal, imag, max, mean, min,
nonzero, prod, ptp, put, ravel, real, repeat, reshape, round, searchsorted, sort, squeeze,
std, sum, swapaxes, take, trace, transpose, var.

Array conversion

ndarray.item(*args) Copy an element of an array to a standard Python scalar and return it.
ndarray.tolist() Return the array as a (possibly nested) list.
ndarray.itemset(*args) Insert scalar into an array (scalar is cast to array’s dtype, if possible)
ndarray.setasflat(arr) Equivalent to a.flat = arr.flat, but is generally more efficient.
ndarray.tostring([order]) Construct a Python string containing the raw data bytes in the array.
ndarray.tofile(fid[, sep, Write array to a file as text or binary (default).
format])
ndarray.dump(file) Dump a pickle of the array to the specified file.
ndarray.dumps() Returns the pickle of the array as a string.
ndarray.astype(t) Copy of the array, cast to a specified type.
ndarray.byteswap(inplace) Swap the bytes of the array elements
ndarray.copy([order]) Return a copy of the array.
ndarray.view([dtype, type]) New view of array with the same data.
ndarray.getfield(dtype, offset) Returns a field of the given array as a certain type.
ndarray.setflags([write, align, Set array flags WRITEABLE, ALIGNED, and UPDATEIFCOPY,
uic]) respectively.
ndarray.fill(value) Fill the array with a scalar value.
ndarray.item(*args)
Copy an element of an array to a standard Python scalar and return it.
Parameters
*args : Arguments (variable number and type)

50 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

• none: in this case, the method only works for arrays with one element (a.size == 1), which
element is copied into a standard Python scalar object and returned.
• int_type: this argument is interpreted as a flat index into the array, specifying which element
to copy and return.
• tuple of int_types: functions as does a single int_type argument, except that the argument is
interpreted as an nd-index into the array.
Returns
z : Standard Python scalar object
A copy of the specified element of the array as a suitable Python scalar

Notes
When the data type of a is longdouble or clongdouble, item() returns a scalar array object because there is no
available Python scalar that would not lose information. Void arrays return a buffer object for item(), unless
fields are defined, in which case a tuple is returned.
item is very similar to a[args], except, instead of an array scalar, a standard Python scalar is returned. This can
be useful for speeding up access to elements of the array and doing arithmetic on elements of the array using
Python’s optimized math.

Examples
>>> x = np.random.randint(9, size=(3, 3))
>>> x
array([[3, 1, 7],
[2, 8, 3],
[8, 5, 3]])
>>> x.item(3)
2
>>> x.item(7)
5
>>> x.item((0, 1))
1
>>> x.item((2, 2))
3

ndarray.tolist()
Return the array as a (possibly nested) list.
Return a copy of the array data as a (nested) Python list. Data items are converted to the nearest compatible
Python type.
Parameters
none :
Returns
y : list
The possibly nested list of array elements.

Notes
The array may be recreated, a = np.array(a.tolist()).

Examples

1.1. The N-dimensional array (ndarray) 51

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a = np.array([1, 2])

>>> a.tolist()
[1, 2]
>>> a = np.array([[1, 2], [3, 4]])
>>> list(a)
[array([1, 2]), array([3, 4])]
>>> a.tolist()
[[1, 2], [3, 4]]

ndarray.itemset(*args)
Insert scalar into an array (scalar is cast to array’s dtype, if possible)
There must be at least 1 argument, and define the last argument as item. Then, a.itemset(*args) is
equivalent to but faster than a[args] = item. The item should be a scalar value and args must select a
single item in the array a.
Parameters
*args : Arguments
If one argument: a scalar, only used in case a is of size 1. If two arguments: the last
argument is the value to be set and must be a scalar, the first argument specifies a single
array element location. It is either an int or a tuple.

Notes
Compared to indexing syntax, itemset provides some speed increase for placing a scalar into a particular location
in an ndarray, if you must do this. However, generally this is discouraged: among other problems, it complicates
the appearance of the code. Also, when using itemset (and item) inside a loop, be sure to assign the methods to
a local variable to avoid the attribute look-up at each loop iteration.

Examples
>>> x = np.random.randint(9, size=(3, 3))
>>> x
array([[3, 1, 7],
[2, 8, 3],
[8, 5, 3]])
>>> x.itemset(4, 0)
>>> x.itemset((2, 2), 9)
>>> x
array([[3, 1, 7],
[2, 0, 3],
[8, 5, 9]])

ndarray.setasflat(arr)
Equivalent to a.flat = arr.flat, but is generally more efficient. This function does not check for overlap, so if arr
and a are viewing the same data with different strides, the results will be unpredictable.
Parameters
arr : array_like
The array to copy into a.

Examples
>>> a = np.arange(2*4).reshape(2,4)[:,:-1]; a
array([[0, 1, 2],
[4, 5, 6]])
>>> b = np.arange(3*3, dtype=’f4’).reshape(3,3).T[::-1,:-1]; b
array([[ 2., 5.],

52 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

[ 1., 4.],
[ 0., 3.]], dtype=float32)
>>> a.setasflat(b)
>>> a
array([[2, 5, 1],
[4, 0, 3]])

ndarray.tostring(order=’C’)
Construct a Python string containing the raw data bytes in the array.
Constructs a Python string showing a copy of the raw contents of data memory. The string can be produced
in either ‘C’ or ‘Fortran’, or ‘Any’ order (the default is ‘C’-order). ‘Any’ order means C-order unless the
F_CONTIGUOUS flag in the array is set, in which case it means ‘Fortran’ order.
Parameters
order : {‘C’, ‘F’, None}, optional
Order of the data for multidimensional arrays: C, Fortran, or the same as for the original
array.
Returns
s : str
A Python string exhibiting a copy of a‘s raw data.

Examples
>>> x = np.array([[0, 1], [2, 3]])
>>> x.tostring()
’\x00\x00\x00\x00\x01\x00\x00\x00\x02\x00\x00\x00\x03\x00\x00\x00’
>>> x.tostring(’C’) == x.tostring()
True
>>> x.tostring(’F’)
’\x00\x00\x00\x00\x02\x00\x00\x00\x01\x00\x00\x00\x03\x00\x00\x00’

ndarray.tofile(fid, sep=”“, format=”%s”)

Write array to a file as text or binary (default).
Data is always written in ‘C’ order, independent of the order of a. The data produced by this method can be
recovered using the function fromfile().
Parameters
fid : file or str
An open file object, or a string containing a filename.
sep : str
Separator between array items for text output. If “” (empty), a binary file is written,
equivalent to file.write(a.tostring()).
format : str
Format string for text file output. Each entry in the array is formatted to text by first
converting it to the closest Python type, and then using “format” % item.

Notes
This is a convenience function for quick storage of array data. Information on endianness and precision is lost,
so this method is not a good choice for files intended to archive data or transport data between machines with
different endianness. Some of these problems can be overcome by outputting the data as text files, at the expense
of speed and file size.

1.1. The N-dimensional array (ndarray) 53

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.dump(file)
Dump a pickle of the array to the specified file. The array can be read back with pickle.load or numpy.load.
Parameters
file : str
A string naming the dump file.
ndarray.dumps()
Returns the pickle of the array as a string. pickle.loads or numpy.loads will convert the string back to an array.
Parameters
None :
ndarray.astype(t)
Copy of the array, cast to a specified type.
Parameters
t : str or dtype
Typecode or data-type to which the array is cast.
Raises
ComplexWarning : :
When casting from complex to float or int. To avoid this, one should use
a.real.astype(t).

Examples
>>> x = np.array([1, 2, 2.5])
>>> x
array([ 1. , 2. , 2.5])

>>> x.astype(int)
array([1, 2, 2])

ndarray.byteswap(inplace)
Swap the bytes of the array elements
Toggle between low-endian and big-endian data representation by returning a byteswapped array, optionally
swapped in-place.
Parameters
inplace: bool, optional :
If True, swap bytes in-place, default is False.
Returns
out: ndarray :
The byteswapped array. If inplace is True, this is a view to self.

Examples
>>> A = np.array([1, 256, 8755], dtype=np.int16)
>>> map(hex, A)
[’0x1’, ’0x100’, ’0x2233’]
>>> A.byteswap(True)
array([ 256, 1, 13090], dtype=int16)
>>> map(hex, A)
[’0x100’, ’0x1’, ’0x3322’]

54 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Arrays of strings are not swapped

>>> A = np.array([’ceg’, ’fac’])
>>> A.byteswap()
array([’ceg’, ’fac’],
dtype=’|S3’)

ndarray.copy(order=’C’)
Return a copy of the array.
Parameters
order : {‘C’, ‘F’, ‘A’}, optional
By default, the result is stored in C-contiguous (row-major) order in memory. If order is
F, the result has ‘Fortran’ (column-major) order. If order is ‘A’ (‘Any’), then the result
has the same order as the input.

Examples
>>> x = np.array([[1,2,3],[4,5,6]], order=’F’)

>>> y = x.copy()

>>> x.fill(0)

>>> x
array([[0, 0, 0],
[0, 0, 0]])

>>> y
array([[1, 2, 3],
[4, 5, 6]])

>>> y.flags[’C_CONTIGUOUS’]
True

ndarray.view(dtype=None, type=None)
New view of array with the same data.
Parameters
dtype : data-type, optional
Data-type descriptor of the returned view, e.g., float32 or int16. The default, None,
results in the view having the same data-type as a.
type : Python type, optional
Type of the returned view, e.g., ndarray or matrix. Again, the default None results in
type preservation.

Notes
a.view() is used two different ways:
a.view(some_dtype) or a.view(dtype=some_dtype) constructs a view of the array’s memory with
a different data-type. This can cause a reinterpretation of the bytes of memory.
a.view(ndarray_subclass) or a.view(type=ndarray_subclass) just returns an instance of
ndarray_subclass that looks at the same array (same shape, dtype, etc.) This does not cause a reinterpretation
of the memory.

1.1. The N-dimensional array (ndarray) 55

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> x = np.array([(1, 2)], dtype=[(’a’, np.int8), (’b’, np.int8)])

Viewing array data using a different type and dtype:

>>> y = x.view(dtype=np.int16, type=np.matrix)
>>> y
matrix([[513]], dtype=int16)
>>> print type(y)
<class ’numpy.matrixlib.defmatrix.matrix’>

Creating a view on a structured array so it can be used in calculations

>>> x = np.array([(1, 2),(3,4)], dtype=[(’a’, np.int8), (’b’, np.int8)])
>>> xv = x.view(dtype=np.int8).reshape(-1,2)
>>> xv
array([[1, 2],
[3, 4]], dtype=int8)
>>> xv.mean(0)
array([ 2., 3.])

Making changes to the view changes the underlying array

>>> xv[0,1] = 20
>>> print x
[(1, 20) (3, 4)]

Using a view to convert an array to a record array:

>>> z = x.view(np.recarray)
>>> z.a
array([1], dtype=int8)

Views share data:

>>> x[0] = (9, 10)
>>> z[0]
(9, 10)

ndarray.getfield(dtype, offset)
Returns a field of the given array as a certain type.
A field is a view of the array data with each itemsize determined by the given type and the offset into the current
array, i.e. from offset * dtype.itemsize to (offset+1) * dtype.itemsize.
Parameters
dtype : str
String denoting the data type of the field.
offset : int
Number of dtype.itemsize‘s to skip before beginning the element view.

Examples
>>> x = np.diag([1.+1.j]*2)
>>> x
array([[ 1.+1.j, 0.+0.j],
[ 0.+0.j, 1.+1.j]])

56 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x.dtype
dtype(’complex128’)

>>> x.getfield(’complex64’, 0) # Note how this != x

array([[ 0.+1.875j, 0.+0.j ],
[ 0.+0.j , 0.+1.875j]], dtype=complex64)

>>> x.getfield(’complex64’,1) # Note how different this is than x

array([[ 0. +5.87173204e-39j, 0. +0.00000000e+00j],
[ 0. +0.00000000e+00j, 0. +5.87173204e-39j]], dtype=complex64)

>>> x.getfield(’complex128’, 0) # == x
array([[ 1.+1.j, 0.+0.j],
[ 0.+0.j, 1.+1.j]])

If the argument dtype is the same as x.dtype, then offset != 0 raises a ValueError:
>>> x.getfield(’complex128’, 1)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ValueError: Need 0 <= offset <= 0 for requested type but received offset = 1

>>> x.getfield(’float64’, 0)
array([[ 1., 0.],
[ 0., 1.]])

>>> x.getfield(’float64’, 1)
array([[ 1.77658241e-307, 0.00000000e+000],
[ 0.00000000e+000, 1.77658241e-307]])

ndarray.setflags(write=None, align=None, uic=None)

Set array flags WRITEABLE, ALIGNED, and UPDATEIFCOPY, respectively.
These Boolean-valued flags affect how numpy interprets the memory area used by a (see Notes below). The
ALIGNED flag can only be set to True if the data is actually aligned according to the type. The UPDATEIF-
COPY flag can never be set to True. The flag WRITEABLE can only be set to True if the array owns its own
memory, or the ultimate owner of the memory exposes a writeable buffer interface, or is a string. (The exception
for string is made so that unpickling can be done without copying memory.)
Parameters
write : bool, optional
Describes whether or not a can be written to.
align : bool, optional
Describes whether or not a is aligned properly for its type.
uic : bool, optional
Describes whether or not a is a copy of another “base” array.

Notes
Array flags provide information about how the memory area used for the array is to be interpreted. There are
6 Boolean flags in use, only three of which can be changed by the user: UPDATEIFCOPY, WRITEABLE, and
ALIGNED.
WRITEABLE (W) the data area can be written to;
ALIGNED (A) the data and strides are aligned appropriately for the hardware (as determined by the compiler);

1.1. The N-dimensional array (ndarray) 57

NumPy Reference, Release 2.0.0.dev-ac3cba3

UPDATEIFCOPY (U) this array is a copy of some other array (referenced by .base). When this array is deallo-
cated, the base array will be updated with the contents of this array.
All flags can be accessed using their first (upper case) letter as well as the full name.

Examples
>>> y
array([[3, 1, 7],
[2, 0, 0],
[8, 5, 9]])
>>> y.flags
C_CONTIGUOUS : True
F_CONTIGUOUS : False
OWNDATA : True
WRITEABLE : True
ALIGNED : True
UPDATEIFCOPY : False
>>> y.setflags(write=0, align=0)
>>> y.flags
C_CONTIGUOUS : True
F_CONTIGUOUS : False
OWNDATA : True
WRITEABLE : False
ALIGNED : False
UPDATEIFCOPY : False
>>> y.setflags(uic=1)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ValueError: cannot set UPDATEIFCOPY flag to True

ndarray.fill(value)
Fill the array with a scalar value.
Parameters
value : scalar
All elements of a will be assigned this value.

Examples
>>> a = np.array([1, 2])
>>> a.fill(0)
>>> a
array([0, 0])
>>> a = np.empty(2)
>>> a.fill(1)
>>> a
array([ 1., 1.])

Shape manipulation

For reshape, resize, and transpose, the single tuple argument may be replaced with n integers which will be interpreted
as an n-tuple.

58 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.reshape(shape[, order]) Returns an array containing the same data with a new shape.
ndarray.resize(new_shape[, refcheck]) Change shape and size of array in-place.
ndarray.transpose(*axes) Returns a view of the array with axes transposed.
ndarray.swapaxes(axis1, axis2) Return a view of the array with axis1 and axis2 interchanged.
ndarray.flatten([order]) Return a copy of the array collapsed into one dimension.
ndarray.ravel() Return a flattened array.
ndarray.squeeze() Remove single-dimensional entries from the shape of a.
ndarray.reshape(shape, order=’C’)
Returns an array containing the same data with a new shape.
Refer to numpy.reshape for full documentation.
See Also:

numpy.reshape
equivalent function

ndarray.resize(new_shape, refcheck=True)
Change shape and size of array in-place.
Parameters
new_shape : tuple of ints, or n ints
Shape of resized array.
refcheck : bool, optional
If False, reference count will not be checked. Default is True.
Returns
None :
Raises
ValueError :
If a does not own its own data or references or views to it exist, and the data memory
must be changed.
SystemError :
If the order keyword argument is specified. This behaviour is a bug in NumPy.
See Also:

resize
Return a new array with the specified shape.

Notes
This reallocates space for the data area if necessary.
Only contiguous arrays (data elements consecutive in memory) can be resized.
The purpose of the reference count check is to make sure you do not use this array as a buffer for another Python
object and then reallocate the memory. However, reference counts can increase in other ways so if you are sure
that you have not shared the memory for this array with another Python object, then you may safely set refcheck
to False.

1.1. The N-dimensional array (ndarray) 59

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
Shrinking an array: array is flattened (in the order that the data are stored in memory), resized, and reshaped:
>>> a = np.array([[0, 1], [2, 3]], order=’C’)
>>> a.resize((2, 1))
>>> a
array([[0],
[1]])

>>> a = np.array([[0, 1], [2, 3]], order=’F’)

>>> a.resize((2, 1))
>>> a
array([[0],
[2]])

Enlarging an array: as above, but missing entries are filled with zeros:
>>> b = np.array([[0, 1], [2, 3]])
>>> b.resize(2, 3) # new_shape parameter doesn’t have to be a tuple
>>> b
array([[0, 1, 2],
[3, 0, 0]])

Referencing an array prevents resizing...

>>> c = a
>>> a.resize((1, 1))
...
ValueError: cannot resize an array that has been referenced ...

Unless refcheck is False:

>>> a.resize((1, 1), refcheck=False)
>>> a
array([[0]])
>>> c
array([[0]])

ndarray.transpose(*axes)
Returns a view of the array with axes transposed.
For a 1-D array, this has no effect. (To change between column and row vectors, first cast the 1-D array into a ma-
trix object.) For a 2-D array, this is the usual matrix transpose. For an n-D array, if axes are given, their order in-
dicates how the axes are permuted (see Examples). If axes are not provided and a.shape = (i[0], i[1],
... i[n-2], i[n-1]), then a.transpose().shape = (i[n-1], i[n-2], ... i[1],
i[0]).
Parameters
axes : None, tuple of ints, or n ints
• None or no argument: reverses the order of the axes.
• tuple of ints: i in the j-th place in the tuple means a‘s i-th axis becomes a.transpose()‘s j-th
axis.
• n ints: same as an n-tuple of the same ints (this form is intended simply as a “convenience”
alternative to the tuple form)
Returns
out : ndarray

60 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

View of a, with axes suitably permuted.

See Also:

ndarray.T
Array property returning the array transposed.

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> a
array([[1, 2],
[3, 4]])
>>> a.transpose()
array([[1, 3],
[2, 4]])
>>> a.transpose((1, 0))
array([[1, 3],
[2, 4]])
>>> a.transpose(1, 0)
array([[1, 3],
[2, 4]])

ndarray.swapaxes(axis1, axis2)
Return a view of the array with axis1 and axis2 interchanged.
Refer to numpy.swapaxes for full documentation.
See Also:

numpy.swapaxes
equivalent function

ndarray.flatten(order=’C’)
Return a copy of the array collapsed into one dimension.
Parameters
order : {‘C’, ‘F’, ‘A’}, optional
Whether to flatten in C (row-major), Fortran (column-major) order, or preserve the
C/Fortran ordering from a. The default is ‘C’.
Returns
y : ndarray
A copy of the input array, flattened to one dimension.
See Also:

ravel
Return a flattened array.
flat
A 1-D flat iterator over the array.

Examples
>>> a = np.array([[1,2], [3,4]])
>>> a.flatten()
array([1, 2, 3, 4])

1.1. The N-dimensional array (ndarray) 61

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a.flatten(’F’)
array([1, 3, 2, 4])

ndarray.ravel([order ])
Return a flattened array.
Refer to numpy.ravel for full documentation.
See Also:

numpy.ravel
equivalent function
ndarray.flat
a flat iterator on the array.

ndarray.squeeze()
Remove single-dimensional entries from the shape of a.
Refer to numpy.squeeze for full documentation.
See Also:

numpy.squeeze
equivalent function

Item selection and manipulation

For array methods that take an axis keyword, it defaults to None. If axis is None, then the array is treated as a 1-D
array. Any other value for axis represents the dimension along which the operation should proceed.
ndarray.take(indices[, axis, out, Return an array formed from the elements of a at the given indices.
mode])
ndarray.put(indices, values[, mode]) Set a.flat[n] = values[n] for all n in indices.
ndarray.repeat(repeats[, axis]) Repeat elements of an array.
ndarray.choose(choices[, out, Use an index array to construct a new array from a set of choices.
mode])
ndarray.sort([axis, kind, order]) Sort an array, in-place.
ndarray.argsort([axis, kind, order]) Returns the indices that would sort this array.
ndarray.searchsorted(v[, side]) Find indices where elements of v should be inserted in a to maintain
order.
ndarray.nonzero() Return the indices of the elements that are non-zero.
ndarray.compress(condition[, axis, Return selected slices of this array along given axis.
out])
ndarray.diagonal([offset, axis1, Return specified diagonals.
axis2])
ndarray.take(indices, axis=None, out=None, mode=’raise’)
Return an array formed from the elements of a at the given indices.
Refer to numpy.take for full documentation.
See Also:

numpy.take
equivalent function

62 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.put(indices, values, mode=’raise’)

Set a.flat[n] = values[n] for all n in indices.
Refer to numpy.put for full documentation.
See Also:

numpy.put
equivalent function

ndarray.repeat(repeats, axis=None)
Repeat elements of an array.
Refer to numpy.repeat for full documentation.
See Also:

numpy.repeat
equivalent function

ndarray.choose(choices, out=None, mode=’raise’)

Use an index array to construct a new array from a set of choices.
Refer to numpy.choose for full documentation.
See Also:

numpy.choose
equivalent function

ndarray.sort(axis=-1, kind=’quicksort’, order=None)

Sort an array, in-place.
Parameters
axis : int, optional
Axis along which to sort. Default is -1, which means sort along the last axis.
kind : {‘quicksort’, ‘mergesort’, ‘heapsort’}, optional
Sorting algorithm. Default is ‘quicksort’.
order : list, optional
When a is an array with fields defined, this argument specifies which fields to compare
first, second, etc. Not all fields need be specified.
See Also:

numpy.sort
Return a sorted copy of an array.
argsort
Indirect sort.
lexsort
Indirect stable sort on multiple keys.
searchsorted
Find elements in sorted array.

1.1. The N-dimensional array (ndarray) 63

NumPy Reference, Release 2.0.0.dev-ac3cba3

Notes
See sort for notes on the different sorting algorithms.

Examples
>>> a = np.array([[1,4], [3,1]])
>>> a.sort(axis=1)
>>> a
array([[1, 4],
[1, 3]])
>>> a.sort(axis=0)
>>> a
array([[1, 3],
[1, 4]])

Use the order keyword to specify a field to use when sorting a structured array:
>>> a = np.array([(’a’, 2), (’c’, 1)], dtype=[(’x’, ’S1’), (’y’, int)])
>>> a.sort(order=’y’)
>>> a
array([(’c’, 1), (’a’, 2)],
dtype=[(’x’, ’|S1’), (’y’, ’<i4’)])

ndarray.argsort(axis=-1, kind=’quicksort’, order=None)

Returns the indices that would sort this array.
Refer to numpy.argsort for full documentation.
See Also:

numpy.argsort
equivalent function

ndarray.searchsorted(v, side=’left’)
Find indices where elements of v should be inserted in a to maintain order.
For full documentation, see numpy.searchsorted
See Also:

numpy.searchsorted
equivalent function

ndarray.nonzero()
Return the indices of the elements that are non-zero.
Refer to numpy.nonzero for full documentation.
See Also:

numpy.nonzero
equivalent function

ndarray.compress(condition, axis=None, out=None)

Return selected slices of this array along given axis.
Refer to numpy.compress for full documentation.
See Also:

64 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.compress
equivalent function

ndarray.diagonal(offset=0, axis1=0, axis2=1)

Return specified diagonals.
Refer to numpy.diagonal for full documentation.
See Also:

numpy.diagonal
equivalent function

Calculation

Many of these methods take an argument named axis. In such cases,

• If axis is None (the default), the array is treated as a 1-D array and the operation is performed over the entire
array. This behavior is also the default if self is a 0-dimensional array or array scalar. (An array scalar is
an instance of the types/classes float32, float64, etc., whereas a 0-dimensional array is an ndarray instance
containing precisely one array scalar.)
• If axis is an integer, then the operation is done over the given axis (for each 1-D subarray that can be created
along the given axis).
Example of the axis argument
A 3-dimensional array of size 3 x 3 x 3, summed over each of its three axes
>>> x
array([[[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8]],
[[ 9, 10, 11],
[12, 13, 14],
[15, 16, 17]],
[[18, 19, 20],
[21, 22, 23],
[24, 25, 26]]])
>>> x.sum(axis=0)
array([[27, 30, 33],
[36, 39, 42],
[45, 48, 51]])
>>> # for sum, axis is the first keyword, so we may omit it,
>>> # specifying only its value
>>> x.sum(0), x.sum(1), x.sum(2)
(array([[27, 30, 33],
[36, 39, 42],
[45, 48, 51]]),
array([[ 9, 12, 15],
[36, 39, 42],
[63, 66, 69]]),
array([[ 3, 12, 21],
[30, 39, 48],
[57, 66, 75]]))

The parameter dtype specifies the data type over which a reduction operation (like summing) should take place. The
default reduce data type is the same as the data type of self. To avoid overflow, it can be useful to perform the reduction
using a larger data type.

1.1. The N-dimensional array (ndarray) 65

NumPy Reference, Release 2.0.0.dev-ac3cba3

For several methods, an optional out argument can also be provided and the result will be placed into the output array
given. The out argument must be an ndarray and have the same number of elements. It can have a different data
type in which case casting will be performed.
ndarray.argmax([axis, out]) Return indices of the maximum values along the given axis.
ndarray.min([axis, out]) Return the minimum along a given axis.
ndarray.argmin([axis, out]) Return indices of the minimum values along the given axis of a.
ndarray.ptp([axis, out]) Peak to peak (maximum - minimum) value along a given axis.
ndarray.clip(a_min, a_max[, out]) Return an array whose values are limited to [a_min,
a_max].
ndarray.conj() Complex-conjugate all elements.
ndarray.round([decimals, out]) Return a with each element rounded to the given number of
decimals.
ndarray.trace([offset, axis1, axis2, dtype, Return the sum along diagonals of the array.
out])
ndarray.sum([axis, dtype, out]) Return the sum of the array elements over the given axis.
ndarray.cumsum([axis, dtype, out]) Return the cumulative sum of the elements along the given
axis.
ndarray.mean([axis, dtype, out]) Returns the average of the array elements along given axis.
ndarray.var([axis, dtype, out, ddof]) Returns the variance of the array elements, along given axis.
ndarray.std([axis, dtype, out, ddof]) Returns the standard deviation of the array elements along
given axis.
ndarray.prod([axis, dtype, out]) Return the product of the array elements over the given axis
ndarray.cumprod([axis, dtype, out]) Return the cumulative product of the elements along the given
axis.
ndarray.all([axis, out]) Returns True if all elements evaluate to True.
ndarray.any([axis, out]) Returns True if any of the elements of a evaluate to True.
ndarray.argmax(axis=None, out=None)
Return indices of the maximum values along the given axis.
Refer to numpy.argmax for full documentation.
See Also:

numpy.argmax
equivalent function

ndarray.min(axis=None, out=None)
Return the minimum along a given axis.
Refer to numpy.amin for full documentation.
See Also:

numpy.amin
equivalent function

ndarray.argmin(axis=None, out=None)
Return indices of the minimum values along the given axis of a.
Refer to numpy.argmin for detailed documentation.
See Also:

numpy.argmin
equivalent function

66 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.ptp(axis=None, out=None)
Peak to peak (maximum - minimum) value along a given axis.
Refer to numpy.ptp for full documentation.
See Also:

numpy.ptp
equivalent function

ndarray.clip(a_min, a_max, out=None)

Return an array whose values are limited to [a_min, a_max].
Refer to numpy.clip for full documentation.
See Also:

numpy.clip
equivalent function

ndarray.conj()
Complex-conjugate all elements.
Refer to numpy.conjugate for full documentation.
See Also:

numpy.conjugate
equivalent function

ndarray.round(decimals=0, out=None)
Return a with each element rounded to the given number of decimals.
Refer to numpy.around for full documentation.
See Also:

numpy.around
equivalent function

ndarray.trace(offset=0, axis1=0, axis2=1, dtype=None, out=None)

Return the sum along diagonals of the array.
Refer to numpy.trace for full documentation.
See Also:

numpy.trace
equivalent function

ndarray.sum(axis=None, dtype=None, out=None)

Return the sum of the array elements over the given axis.
Refer to numpy.sum for full documentation.
See Also:

numpy.sum
equivalent function

1.1. The N-dimensional array (ndarray) 67

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.cumsum(axis=None, dtype=None, out=None)

Return the cumulative sum of the elements along the given axis.
Refer to numpy.cumsum for full documentation.
See Also:

numpy.cumsum
equivalent function

ndarray.mean(axis=None, dtype=None, out=None)

Returns the average of the array elements along given axis.
Refer to numpy.mean for full documentation.
See Also:

numpy.mean
equivalent function

ndarray.var(axis=None, dtype=None, out=None, ddof=0)

Returns the variance of the array elements, along given axis.
Refer to numpy.var for full documentation.
See Also:

numpy.var
equivalent function

ndarray.std(axis=None, dtype=None, out=None, ddof=0)

Returns the standard deviation of the array elements along given axis.
Refer to numpy.std for full documentation.
See Also:

numpy.std
equivalent function

ndarray.prod(axis=None, dtype=None, out=None)

Return the product of the array elements over the given axis
Refer to numpy.prod for full documentation.
See Also:

numpy.prod
equivalent function

ndarray.cumprod(axis=None, dtype=None, out=None)

Return the cumulative product of the elements along the given axis.
Refer to numpy.cumprod for full documentation.
See Also:

numpy.cumprod
equivalent function

68 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.all(axis=None, out=None)
Returns True if all elements evaluate to True.
Refer to numpy.all for full documentation.
See Also:

numpy.all
equivalent function

ndarray.any(axis=None, out=None)
Returns True if any of the elements of a evaluate to True.
Refer to numpy.any for full documentation.
See Also:

numpy.any
equivalent function

1.1.6 Arithmetic and comparison operations

Arithmetic and comparison operations on ndarrays are defined as element-wise operations, and generally yield
ndarray objects as results.
Each of the arithmetic operations (+, -, *, /, //, %, divmod(), ** or pow(), <<, >>, &, ^, |, ~) and the
comparisons (==, <, >, <=, >=, !=) is equivalent to the corresponding universal function (or ufunc for short) in
Numpy. For more information, see the section on Universal Functions.
Comparison operators:
ndarray.__lt__ x.__lt__(y) <==> x<y
ndarray.__le__ x.__le__(y) <==> x<=y
ndarray.__gt__ x.__gt__(y) <==> x>y
ndarray.__ge__ x.__ge__(y) <==> x>=y
ndarray.__eq__ x.__eq__(y) <==> x==y
ndarray.__ne__ x.__ne__(y) <==> x!=y
ndarray.__lt__()
x.__lt__(y) <==> x<y
ndarray.__le__()
x.__le__(y) <==> x<=y
ndarray.__gt__()
x.__gt__(y) <==> x>y
ndarray.__ge__()
x.__ge__(y) <==> x>=y
ndarray.__eq__()
x.__eq__(y) <==> x==y
ndarray.__ne__()
x.__ne__(y) <==> x!=y
Truth value of an array (bool):
ndarray.__nonzero__ x.__nonzero__() <==> x != 0

1.1. The N-dimensional array (ndarray) 69

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.__nonzero__()
x.__nonzero__() <==> x != 0
Note: Truth-value testing of an array invokes ndarray.__nonzero__, which raises an error if the number of
elements in the the array is larger than 1, because the truth value of such arrays is ambiguous. Use .any() and
.all() instead to be clear about what is meant in such cases. (If the number of elements is 0, the array evaluates to
False.)
Unary operations:
ndarray.__neg__ x.__neg__() <==> -x
ndarray.__pos__ x.__pos__() <==> +x
ndarray.__abs__(x)
ndarray.__invert__ x.__invert__() <==> ~x
ndarray.__neg__()
x.__neg__() <==> -x
ndarray.__pos__()
x.__pos__() <==> +x
ndarray.__abs__() <==> abs(x)

ndarray.__invert__()
x.__invert__() <==> ~x
Arithmetic:
ndarray.__add__ x.__add__(y) <==> x+y
ndarray.__sub__ x.__sub__(y) <==> x-y
ndarray.__mul__ x.__mul__(y) <==> x*y
ndarray.__div__ x.__div__(y) <==> x/y
ndarray.__truediv__ x.__truediv__(y) <==> x/y
ndarray.__floordiv__ x.__floordiv__(y) <==> x//y
ndarray.__mod__ x.__mod__(y) <==> x%y
ndarray.__divmod__(x, y)
ndarray.__pow__(x)
ndarray.__lshift__ x.__lshift__(y) <==> x<<y
ndarray.__rshift__ x.__rshift__(y) <==> x>>y
ndarray.__and__ x.__and__(y) <==> x&y
ndarray.__or__ x.__or__(y) <==> x|y
ndarray.__xor__ x.__xor__(y) <==> x^y
ndarray.__add__()
x.__add__(y) <==> x+y
ndarray.__sub__()
x.__sub__(y) <==> x-y
ndarray.__mul__()
x.__mul__(y) <==> x*y
ndarray.__div__()
x.__div__(y) <==> x/y
ndarray.__truediv__()
x.__truediv__(y) <==> x/y
ndarray.__floordiv__()
x.__floordiv__(y) <==> x//y

70 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.__mod__()
x.__mod__(y) <==> x%y
ndarray.__divmod__(y) <==> divmod(x, y)

ndarray.pow(y[, z ], ) <==> pow(x, y[, z ])

ndarray.__lshift__()
x.__lshift__(y) <==> x<<y
ndarray.__rshift__()
x.__rshift__(y) <==> x>>y
ndarray.__and__()
x.__and__(y) <==> x&y
ndarray.__or__()
x.__or__(y) <==> x|y
ndarray.__xor__()
x.__xor__(y) <==> x^y
Note:
• Any third argument to pow is silently ignored, as the underlying ufunc takes only two arguments.
• The three division operators are all defined; div is active by default, truediv is active when __future__
division is in effect.
• Because ndarray is a built-in type (written in C), the __r{op}__ special methods are not directly defined.
• The functions called to implement many arithmetic special methods for arrays can be modified using
set_numeric_ops.
Arithmetic, in-place:
ndarray.__iadd__ x.__iadd__(y) <==> x+y
ndarray.__isub__ x.__isub__(y) <==> x-y
ndarray.__imul__ x.__imul__(y) <==> x*y
ndarray.__idiv__ x.__idiv__(y) <==> x/y
ndarray.__itruediv__ x.__itruediv__(y) <==> x/y
ndarray.__ifloordiv__ x.__ifloordiv__(y) <==> x//y
ndarray.__imod__ x.__imod__(y) <==> x%y
ndarray.__ipow__ x.__ipow__(y) <==> x**y
ndarray.__ilshift__ x.__ilshift__(y) <==> x<<y
ndarray.__irshift__ x.__irshift__(y) <==> x>>y
ndarray.__iand__ x.__iand__(y) <==> x&y
ndarray.__ior__ x.__ior__(y) <==> x|y
ndarray.__ixor__ x.__ixor__(y) <==> x^y
ndarray.__iadd__()
x.__iadd__(y) <==> x+y
ndarray.__isub__()
x.__isub__(y) <==> x-y
ndarray.__imul__()
x.__imul__(y) <==> x*y
ndarray.__idiv__()
x.__idiv__(y) <==> x/y

1.1. The N-dimensional array (ndarray) 71

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.__itruediv__()
x.__itruediv__(y) <==> x/y
ndarray.__ifloordiv__()
x.__ifloordiv__(y) <==> x//y
ndarray.__imod__()
x.__imod__(y) <==> x%y
ndarray.__ipow__()
x.__ipow__(y) <==> x**y
ndarray.__ilshift__()
x.__ilshift__(y) <==> x<<y
ndarray.__irshift__()
x.__irshift__(y) <==> x>>y
ndarray.__iand__()
x.__iand__(y) <==> x&y
ndarray.__ior__()
x.__ior__(y) <==> x|y
ndarray.__ixor__()
x.__ixor__(y) <==> x^y

Warning: In place operations will perform the calculation using the precision decided by the data type of the
two operands, but will silently downcast the result (if necessary) so it can fit back into the array. Therefore,
for mixed precision calculations, A {op}= B can be different than A = A {op} B. For example, suppose
a = ones((3,3)). Then, a += 3j is different than a = a + 3j: while they both perform the same
computation, a += 3 casts the result to fit back in a, whereas a = a + 3j re-binds the name a to the result.

1.1.7 Special methods

For standard library functions:

ndarray.__copy__() Return a copy of the array.
ndarray.__deepcopy__ a.__deepcopy__() -> Deep copy of array.
ndarray.__reduce__() For pickling.
ndarray.__setstate__(version, shape, dtype, ...) For unpickling.

ndarray.__copy__([order ])
Return a copy of the array.
Parameters
order : {‘C’, ‘F’, ‘A’}, optional
If order is ‘C’ (False) then the result is contiguous (default). If order is ‘Fortran’ (True)
then the result has fortran order. If order is ‘Any’ (None) then the result has fortran
order only if the array already is in fortran order.
ndarray.__deepcopy__()
a.__deepcopy__() -> Deep copy of array.
Used if copy.deepcopy is called on an array.
ndarray.__reduce__()
For pickling.

72 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.setstate(version, shape, dtype, isfortran, rawdata)

For unpickling.
Parameters
version : int
optional pickle version. If omitted defaults to 0.
shape : tuple
dtype : data-type
isFortran : bool
rawdata : string or list
a binary string with the data (or a list if ‘a’ is an object array)
Basic customization:
ndarray.__new__
ndarray.__array__ a.__array__(|dtype) -> reference if type unchanged, copy otherwise.
ndarray.__array_wrap__ a.__array_wrap__(obj) -> Object of same type as ndarray object a.
ndarray.__array__()
a.__array__(|dtype) -> reference if type unchanged, copy otherwise.
Returns either a new reference to self if dtype is not given or a new array of provided data type if dtype is
different from the current dtype of the array.
ndarray.__array_wrap__()
a.__array_wrap__(obj) -> Object of same type as ndarray object a.
Container customization: (see Indexing)
ndarray.__len__(x)
ndarray.__getitem__ x.__getitem__(y) <==> x[y]
ndarray.__setitem__ x.__setitem__(i, y) <==> x[i]=y
ndarray.__getslice__ x.__getslice__(i, j) <==> x[i:j]
ndarray.__setslice__ x.__setslice__(i, j, y) <==> x[i:j]=y
ndarray.__contains__ x.__contains__(y) <==> y in x
ndarray.__len__() <==> len(x)

ndarray.__getitem__()
x.__getitem__(y) <==> x[y]
ndarray.__setitem__()
x.__setitem__(i, y) <==> x[i]=y
ndarray.__getslice__()
x.__getslice__(i, j) <==> x[i:j]
Use of negative indices is not supported.
ndarray.__setslice__()
x.__setslice__(i, j, y) <==> x[i:j]=y
Use of negative indices is not supported.
ndarray.__contains__()
x.__contains__(y) <==> y in x

1.1. The N-dimensional array (ndarray) 73

NumPy Reference, Release 2.0.0.dev-ac3cba3

Conversion; the operations complex, int, long, float, oct, and hex. They work only on arrays that have one
element in them and return the appropriate scalar.
ndarray.__int__(x)
ndarray.__long__(x)
ndarray.__float__(x)
ndarray.__oct__(x)
ndarray.__hex__(x)
ndarray.__int__() <==> int(x)

ndarray.long() <==> long(x)

ndarray.float() <==> float(x)

ndarray.oct() <==> oct(x)

ndarray.hex() <==> hex(x)

String representations:
ndarray.__str__(x)
ndarray.__repr__(x)
ndarray.__str__() <==> str(x)

ndarray.repr() <==> repr(x)

1.2 Scalars

Python defines only one type of a particular data class (there is only one integer type, one floating-point type, etc.).
This can be convenient in applications that don’t need to be concerned with all the ways data can be represented in a
computer. For scientific computing, however, more control is often needed.
In NumPy, there are 24 new fundamental Python types to describe different types of scalars. These type descriptors
are mostly based on the types available in the C language that CPython is written in, with several additional types
compatible with Python’s types.
Array scalars have the same attributes and methods as ndarrays. 1 This allows one to treat items of an array partly
on the same footing as arrays, smoothing out rough edges that result when mixing scalar and array operations.
Array scalars live in a hierarchy (see the Figure below) of data types. They can be detected using the hierarchy:
For example, isinstance(val, np.generic) will return True if val is an array scalar object. Alternatively,
what kind of array scalar is present can be determined using other members of the data type hierarchy. Thus, for
example isinstance(val, np.complexfloating) will return True if val is a complex valued type, while
isinstance(val, np.flexible) will return true if val is one of the flexible itemsize array types (string,
unicode, void).
1 However, array scalars are immutable, so none of the array scalar attributes are settable.

74 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Figure 1.2: Figure: Hierarchy of type objects representing the array data types. Not shown are the two integer types
intp and uintp which just point to the integer type that holds a pointer for the platform. All the number types can
be obtained using bit-width names as well.

1.2. Scalars 75
NumPy Reference, Release 2.0.0.dev-ac3cba3

1.2.1 Built-in scalar types

The built-in scalar types are shown below. Along with their (mostly) C-derived names, the integer, float, and complex
data-types are also available using a bit-width convention so that an array of the right size can always be ensured (e.g.
int8, float64, complex128). Two aliases (intp and uintp) pointing to the integer type that is sufficiently
large to hold a C pointer are also provided. The C-like names are associated with character codes, which are shown in
the table. Use of the character codes, however, is discouraged.
Five of the scalar types are essentially equivalent to fundamental Python types and therefore inherit from them as well
as from the generic array scalar type:
Array scalar type Related Python type
int_ IntType
float_ FloatType
complex_ ComplexType
str_ StringType
unicode_ UnicodeType
The bool_ data type is very similar to the Python BooleanType but does not inherit from it because Python’s
BooleanType does not allow itself to be inherited from, and on the C-level the size of the actual bool data is not the
same as a Python Boolean scalar.

Warning: The bool_ type is not a subclass of the int_ type (the bool_ is not even a number type). This is
different than Python’s default implementation of bool as a sub-class of int.

Tip: The default data type in Numpy is float_.

In the tables below, platform? means that the type may not be available on all platforms. Compatibility with
different C or Python types is indicated: two types are compatible if their data is of the same size and interpreted in
the same way.
Booleans:
Type Remarks Character code
bool_ compatible: Python bool ’?’
bool8 8 bits
Integers:
byte compatible: C char ’b’
short compatible: C short ’h’
intc compatible: C int ’i’
int_ compatible: Python int ’l’
longlong compatible: C long long ’q’
intp large enough to fit a pointer ’p’
int8 8 bits
int16 16 bits
int32 32 bits
int64 64 bits
Unsigned integers:

76 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ubyte compatible: C unsigned char ’B’

ushort compatible: C unsigned short ’H’
uintc compatible: C unsigned int ’I’
uint compatible: Python int ’L’
ulonglong compatible: C long long ’Q’
uintp large enough to fit a pointer ’P’
uint8 8 bits
uint16 16 bits
uint32 32 bits
uint64 64 bits
Floating-point numbers:
half ’e’
single compatible: C float ’f’
double compatible: C double
float_ compatible: Python float ’d’
longfloat compatible: C long float ’g’
float16 16 bits
float32 32 bits
float64 64 bits
float96 96 bits, platform?
float128 128 bits, platform?
Complex floating-point numbers:
csingle ’F’
complex_ compatible: Python complex ’D’
clongfloat ’G’
complex64 two 32-bit floats
complex128 two 64-bit floats
complex192 two 96-bit floats, platform?
complex256 two 128-bit floats, platform?
Any Python object:
object_ any Python object ’O’
Note: The data actually stored in object arrays (i.e., arrays having dtype object_) are references to Python objects,
not the objects themselves. Hence, object arrays behave more like usual Python lists, in the sense that their contents
need not be of the same Python type.
The object type is also special because an array containing object_ items does not return an object_ object on
item access, but instead returns the actual object that the array item refers to.
The following data types are flexible. They have no predefined size: the data they describe can be of different length
in different arrays. (In the character codes # is an integer denoting how many elements the data type consists of.)
str_ compatible: Python str ’S#’
unicode_ compatible: Python unicode ’U#’
void ’V#’

Warning: Numeric Compatibility: If you used old typecode characters in your Numeric code (which was never
recommended), you will need to change some of them to the new characters. In particular, the needed changes
are c -> S1, b -> B, 1 -> b, s -> h, w -> H, and u -> I. These changes make the type character
convention more consistent with other Python modules such as the struct module.

1.2. Scalars 77
NumPy Reference, Release 2.0.0.dev-ac3cba3

1.2.2 Attributes

The array scalar objects have an array priority of NPY_SCALAR_PRIORITY (-1,000,000.0). They also do
not (yet) have a ctypes attribute. Otherwise, they share the same attributes as arrays:
generic.flags integer value of flags
generic.shape tuple of array dimensions
generic.strides tuple of bytes steps in each dimension
generic.ndim number of array dimensions
generic.data pointer to start of data
generic.size number of elements in the gentype
generic.itemsize length of one element in bytes
generic.base base object
generic.dtype get array data-descriptor
generic.real real part of scalar
generic.imag imaginary part of scalar
generic.flat a 1-d view of scalar
generic.T transpose
generic.__array_interface__ Array protocol: Python side
generic.__array_struct__ Array protocol: struct
generic.__array_priority__ Array priority.
generic.__array_wrap__ sc.__array_wrap__(obj) return scalar from array
generic.flags
integer value of flags
generic.shape
tuple of array dimensions
generic.strides
tuple of bytes steps in each dimension
generic.ndim
number of array dimensions
generic.data
pointer to start of data
generic.size
number of elements in the gentype
generic.itemsize
length of one element in bytes
generic.base
base object
generic.dtype
get array data-descriptor
generic.real
real part of scalar
generic.imag
imaginary part of scalar
generic.flat
a 1-d view of scalar
generic.T
transpose

78 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

generic.__array_interface__
Array protocol: Python side
generic.__array_struct__
Array protocol: struct
generic.__array_priority__
Array priority.
generic.__array_wrap__()
sc.__array_wrap__(obj) return scalar from array

1.2.3 Indexing

See Also:
Indexing, Data type objects (dtype)
Array scalars can be indexed like 0-dimensional arrays: if x is an array scalar,
• x[()] returns a 0-dimensional ndarray
• x[’field-name’] returns the array scalar in the field field-name. (x can have fields, for example, when it
corresponds to a record data type.)

1.2.4 Methods

Array scalars have exactly the same methods as arrays. The default behavior of these methods is to internally convert
the scalar to an equivalent 0-dimensional array and to call the corresponding array method. In addition, math operations
on array scalars are defined so that the same hardware flags are set and used to interpret the results as for ufunc, so that
the error state used for ufuncs also carries over to the math on array scalars.
The exceptions to the above rules are given below:
generic Base class for numpy scalar types.
generic.__array__ sc.__array__(|type) return 0-dim array
generic.__array_wrap__ sc.__array_wrap__(obj) return scalar from array
generic.squeeze Not implemented (virtual attribute)
generic.byteswap Not implemented (virtual attribute)
generic.__reduce__
generic.__setstate__
generic.setflags Not implemented (virtual attribute)
class numpy.generic
Base class for numpy scalar types.
Class from which most (all?) numpy scalar types are derived. For consistency, exposes the same API as ndarray,
despite many consequent attributes being either “get-only,” or completely irrelevant. This is the class from which
it is strongly suggested users should derive custom scalar types.

Methods

1.2. Scalars 79
NumPy Reference, Release 2.0.0.dev-ac3cba3

Table 1.2 – continued from previous page

astype
byteswap
choose(a, choices[, out, mode]) Construct an array from an index array and a set of arrays to choose from.
clip(a, a_min, a_max[, out]) Clip (limit) the values in an array.
compress(condition, a[, axis, out]) Return selected slices of an array along given axis.
conj() Return the complex conjugate, element-wise.
conjugate() Return the complex conjugate, element-wise.
copy(a) Return an array copy of the given object.
cumprod(a[, axis, dtype, out]) Return the cumulative product of elements along a given axis.
cumsum(a[, axis, dtype, out]) Return the cumulative sum of the elements along a given axis.
diagonal(a[, offset, axis1, axis2]) Return specified diagonals.
dump
dumps
fill
flatten
getfield
item
itemset
max(a[, axis, out]) Return the maximum of an array or maximum along an axis.
mean(a[, axis, dtype, out]) Compute the arithmetic mean along the specified axis.
min(a[, axis, out]) Return the minimum of an array or minimum along an axis.
newbyteorder
nonzero(a) Return the indices of the elements that are non-zero.
prod(a[, axis, dtype, out]) Return the product of array elements over a given axis.
ptp(a[, axis, out]) Range of values (maximum - minimum) along an axis.
put(a, ind, v[, mode]) Replaces specified elements of an array with given values.
ravel(a[, order]) Return a flattened array.
repeat(a, repeats[, axis]) Repeat elements of an array.
reshape(a, newshape[, order]) Gives a new shape to an array without changing its data.
resize(a, new_shape) Return a new array with the specified shape.
round(a[, decimals, out]) Round an array to the given number of decimals.
searchsorted(a, v[, side]) Find indices where elements should be inserted to maintain order.
setfield
setflags
sort(a[, axis, kind, order]) Return a sorted copy of an array.
squeeze(a) Remove single-dimensional entries from the shape of an array.
std(a[, axis, dtype, out, ddof]) Compute the standard deviation along the specified axis.
sum(a[, axis, dtype, out]) Sum of array elements over a given axis.
swapaxes(a, axis1, axis2) Interchange two axes of an array.
take(a, indices[, axis, out, mode]) Take elements from an array along an axis.
tofile
tolist
tostring
trace(a[, offset, axis1, axis2, dtype, out]) Return the sum along diagonals of the array.
transpose(a[, axes]) Permute the dimensions of an array.
var(a[, axis, dtype, out, ddof]) Compute the variance along the specified axis.
view

numpy.all(a, axis=None, out=None)

Test whether all array elements along a given axis evaluate to True.
Parameters
a : array_like

80 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Input array or object that can be converted to an array.

axis : int, optional
Axis along which a logical AND is performed. The default (axis = None) is to perform
a logical AND over a flattened input array. axis may be negative, in which case it counts
from the last to the first axis.
out : ndarray, optional
Alternate output array in which to place the result. It must have the same shape as
the expected output and its type is preserved (e.g., if dtype(out) is float, the result
will consist of 0.0’s and 1.0’s). See doc.ufuncs (Section “Output arguments”) for more
details.
Returns
all : ndarray, bool
A new boolean or array is returned unless out is specified, in which case a reference to
out is returned.
See Also:

ndarray.all
equivalent method
any
Test whether any element along a given axis evaluates to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.all([[True,False],[True,True]])
False

>>> np.all([[True,False],[True,True]], axis=0)

array([ True, False], dtype=bool)

>>> np.all([-1, 4, 5])

True

>>> np.all([1.0, np.nan])

True

>>> o=np.array([False])
>>> z=np.all([-1, 4, 5], out=o)
>>> id(z), id(o), z
(28293632, 28293632, array([ True], dtype=bool))

numpy.any(a, axis=None, out=None)

Test whether any array element along a given axis evaluates to True.
Returns single boolean unless axis is not None
Parameters
a : array_like
Input array or object that can be converted to an array.

1.2. Scalars 81
NumPy Reference, Release 2.0.0.dev-ac3cba3

axis : int, optional

Axis along which a logical OR is performed. The default (axis = None) is to perform
a logical OR over a flattened input array. axis may be negative, in which case it counts
from the last to the first axis.
out : ndarray, optional
Alternate output array in which to place the result. It must have the same shape as the
expected output and its type is preserved (e.g., if it is of type float, then it will remain
so, returning 1.0 for True and 0.0 for False, regardless of the type of a). See doc.ufuncs
(Section “Output arguments”) for details.
Returns
any : bool or ndarray
A new boolean or ndarray is returned unless out is specified, in which case a reference
to out is returned.
See Also:

ndarray.any
equivalent method
all
Test whether all elements along a given axis evaluate to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.any([[True, False], [True, True]])
True

>>> np.any([[True, False], [False, False]], axis=0)

array([ True, False], dtype=bool)

>>> np.any([-1, 0, 5])

True

>>> np.any(np.nan)
True

numpy.argmax(a, axis=None)
Indices of the maximum values along an axis.
Parameters
a : array_like

82 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Input array.
axis : int, optional
By default, the index is into the flattened array, otherwise along the specified axis.
Returns
index_array : ndarray of ints
Array of indices into the array. It has the same shape as a.shape with the dimension
along axis removed.
See Also:
ndarray.argmax, argmin
amax
The maximum value along a given axis.
unravel_index
Convert a flat index into an index tuple.

Notes
In case of multiple occurrences of the maximum values, the indices corresponding to the first occurrence
are returned.

Examples
>>> a = np.arange(6).reshape(2,3)
>>> a
array([[0, 1, 2],
[3, 4, 5]])
>>> np.argmax(a)
5
>>> np.argmax(a, axis=0)
array([1, 1, 1])
>>> np.argmax(a, axis=1)
array([2, 2])

>>> b = np.arange(6)
>>> b[1] = 5
>>> b
array([0, 5, 2, 3, 4, 5])
>>> np.argmax(b) # Only the first occurrence is returned.
1

numpy.argmin(a, axis=None)
Return the indices of the minimum values along an axis.
See Also:

argmax
Similar function. Please refer to numpy.argmax for detailed documentation.

numpy.argsort(a, axis=-1, kind=’quicksort’, order=None)

1.2. Scalars 83
NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
a : array_like
Array to sort.
axis : int or None, optional
Axis along which to sort. The default is -1 (the last axis). If None, the flattened array is
used.
kind : {‘quicksort’, ‘mergesort’, ‘heapsort’}, optional
Sorting algorithm.
order : list, optional
When a is an array with fields defined, this argument specifies which fields to compare
first, second, etc. Not all fields need be specified.
Returns
index_array : ndarray, int
Array of indices that sort a along the specified axis. In other words,
a[index_array] yields a sorted a.
See Also:

sort
Describes sorting algorithms used.
lexsort
Indirect stable sort with multiple keys.
ndarray.sort
Inplace sort.

Notes
See sort for notes on the different sorting algorithms.
As of NumPy 1.4.0 argsort works with real/complex arrays containing nan values. The enhanced sort
order is documented in sort.

Examples
One dimensional array:
>>> x = np.array([3, 1, 2])
>>> np.argsort(x)
array([1, 2, 0])

Two-dimensional array:
>>> x = np.array([[0, 3], [2, 2]])
>>> x
array([[0, 3],
[2, 2]])

>>> np.argsort(x, axis=0)

array([[0, 1],
[1, 0]])

84 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.argsort(x, axis=1)

array([[0, 1],
[0, 1]])

Sorting with keys:

>>> x = np.array([(1, 0), (0, 1)], dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])
>>> x
array([(1, 0), (0, 1)],
dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])

>>> np.argsort(x, order=(’x’,’y’))

array([1, 0])

>>> np.argsort(x, order=(’y’,’x’))

array([0, 1])

numpy.choose(a, choices, out=None, mode=’raise’)

Construct an array from an index array and a set of arrays to choose from.
First of all, if confused or uncertain, definitely look at the Examples - in its full generality,
this function is less simple than it might seem from the following code description (below ndi =
numpy.lib.index_tricks):
np.choose(a,c) == np.array([c[a[I]][I] for I in ndi.ndindex(a.shape)]).
But this omits some subtleties. Here is a fully general summary:
Given an “index” array (a) of integers and a sequence of n arrays (choices), a and each choice array are
first broadcast, as necessary, to arrays of a common shape; calling these Ba and Bchoices[i], i = 0,...,n-1
we have that, necessarily, Ba.shape == Bchoices[i].shape for each i. Then, a new array with
shape Ba.shape is created as follows:
•if mode=raise (the default), then, first of all, each element of a (and thus Ba) must be in the range
[0, n-1]; now, suppose that i (in that range) is the value at the (j0, j1, ..., jm) position in Ba - then the
value at the same position in the new array is the value in Bchoices[i] at that same position;
•if mode=wrap, values in a (and thus Ba) may be any (signed) integer; modular arithmetic is used to
map integers outside the range [0, n-1] back into that range; and then the new array is constructed as
above;
•if mode=clip, values in a (and thus Ba) may be any (signed) integer; negative integers are mapped
to 0; values greater than n-1 are mapped to n-1; and then the new array is constructed as above.

1.2. Scalars 85
NumPy Reference, Release 2.0.0.dev-ac3cba3

Specifies how indices outside [0, n-1] will be treated:

• ‘raise’ : an exception is raised
• ‘wrap’ : value becomes value mod n
• ‘clip’ : values < 0 are mapped to 0, values > n-1 are mapped to n-1
Returns
merged_array : array
The merged result.
Raises
ValueError: shape mismatch :
If a and each choice array are not all broadcastable to the same shape.

See Also:

ndarray.choose
equivalent method

A couple examples illustrating how choose broadcasts:

>>> a = [[1, 0, 1], [0, 1, 0], [1, 0, 1]]
>>> choices = [-10, 10]
>>> np.choose(a, choices)
array([[ 10, -10, 10],
[-10, 10, -10],
[ 10, -10, 10]])

>>> # With thanks to Anne Archibald

86 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

array([[[ 1, 1, 1, 1, 1],
[ 2, 2, 2, 2, 2],
[ 3, 3, 3, 3, 3]],
[[-1, -2, -3, -4, -5],
[-1, -2, -3, -4, -5],
[-1, -2, -3, -4, -5]]])

numpy.clip(a, a_min, a_max, out=None)

Clip (limit) the values in an array.
Given an interval, values outside the interval are clipped to the interval edges. For example, if an interval
of [0, 1] is specified, values smaller than 0 become 0, and values larger than 1 become 1.
Parameters
a : array_like
Array containing elements to clip.
a_min : scalar or array_like
Minimum value.
a_max : scalar or array_like
Maximum value. If a_min or a_max are array_like, then they will be broadcasted to the
shape of a.
out : ndarray, optional
The results will be placed in this array. It may be the input array for in-place clipping.
out must be of the right shape to hold the output. Its type is preserved.
Returns
clipped_array : ndarray
An array with the elements of a, but where values < a_min are replaced with a_min, and
those > a_max with a_max.
See Also:

numpy.doc.ufuncs
Section “Output arguments”

numpy.compress(condition, a, axis=None, out=None)

Return selected slices of an array along given axis.

1.2. Scalars 87
NumPy Reference, Release 2.0.0.dev-ac3cba3

When working along a given axis, a slice along that axis is returned in output for each index where condi-
tion evaluates to True. When working on a 1-D array, compress is equivalent to extract.
Parameters
condition : 1-D array of bools
Array that selects which entries to return. If len(condition) is less than the size of a
along the given axis, then output is truncated to the length of the condition array.
a : array_like
Array from which to extract a part.
axis : int, optional
Axis along which to take slices. If None (default), work on the flattened array.
out : ndarray, optional
Output array. Its type is preserved and it must be of the right shape to hold the output.
Returns
compressed_array : ndarray
A copy of a without the slices along axis for which condition is false.
See Also:
take, choose, diag, diagonal, select
ndarray.compress
Equivalent method.
numpy.doc.ufuncs
Section “Output arguments”

Working on the flattened array does not return slices along an axis but selects elements.
>>> np.compress([False, True], a)
array([2])

numpy.conj(x[, out ])
Return the complex conjugate, element-wise.
The complex conjugate of a complex number is obtained by changing the sign of its imaginary part.

88 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
x : array_like
Input value.
Returns
y : ndarray
The complex conjugate of x, with same dtype as y.

Examples
>>> np.conjugate(1+2j)
(1-2j)

>>> x = np.eye(2) + 1j * np.eye(2)

>>> np.conjugate(x)
array([[ 1.-1.j, 0.-0.j],
[ 0.-0.j, 1.-1.j]])

numpy.copy(a)
Return an array copy of the given object.
Parameters
a : array_like
Input data.
Returns
arr : ndarray
Array interpretation of a.

Notes
This is equivalent to
>>> np.array(a, copy=True)

Examples
Create an array x, with a reference y and a copy z:
>>> x = np.array([1, 2, 3])
>>> y = x
>>> z = np.copy(x)

Note that, when we modify x, y changes, but not z:

>>> x[0] = 10
>>> x[0] == y[0]
True
>>> x[0] == z[0]
False

numpy.cumprod(a, axis=None, dtype=None, out=None)

Return the cumulative product of elements along a given axis.
Parameters
a : array_like
Input array.

1.2. Scalars 89
NumPy Reference, Release 2.0.0.dev-ac3cba3

axis : int, optional

Axis along which the cumulative product is computed. By default the input is flattened.
dtype : dtype, optional
Type of the returned array, as well as of the accumulator in which the elements are
multiplied. If dtype is not specified, it defaults to the dtype of a, unless a has an integer
dtype with a precision less than that of the default platform integer. In that case, the
default platform integer is used instead.
out : ndarray, optional
Alternative output array in which to place the result. It must have the same shape and
buffer length as the expected output but the type of the resulting values will be cast if
necessary.
Returns
cumprod : ndarray
A new array holding the result is returned unless out is specified, in which case a refer-
ence to out is returned.
See Also:

numpy.doc.ufuncs
Section “Output arguments”

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

The cumulative product for each column (i.e., over the rows) of a:
>>> np.cumprod(a, axis=0)
array([[ 1, 2, 3],
[ 4, 10, 18]])

The cumulative product for each row (i.e. over the columns) of a:
>>> np.cumprod(a,axis=1)
array([[ 1, 2, 6],
[ 4, 20, 120]])

numpy.cumsum(a, axis=None, dtype=None, out=None)

Return the cumulative sum of the elements along a given axis.
Parameters
a : array_like
Input array.
axis : int, optional

90 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Axis along which the cumulative sum is computed. The default (None) is to compute
the cumsum over the flattened array.
dtype : dtype, optional
Type of the returned array and of the accumulator in which the elements are summed. If
dtype is not specified, it defaults to the dtype of a, unless a has an integer dtype with a
precision less than that of the default platform integer. In that case, the default platform
integer is used.
out : ndarray, optional
Alternative output array in which to place the result. It must have the same shape and
buffer length as the expected output but the type will be cast if necessary. See doc.ufuncs
(Section “Output arguments”) for more details.
Returns
cumsum_along_axis : ndarray.
A new array holding the result is returned unless out is specified, in which case a ref-
erence to out is returned. The result has the same size as a, and the same shape as a if
axis is not None or a is a 1-d array.
See Also:

sum
Sum array elements.
trapz
Integration of array values using the composite trapezoidal rule.

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

>>> np.cumsum(a,axis=0) # sum over rows for each of the 3 columns

array([[1, 2, 3],
[5, 7, 9]])
>>> np.cumsum(a,axis=1) # sum over columns for each of the 2 rows
array([[ 1, 3, 6],
[ 4, 9, 15]])

numpy.diagonal(a, offset=0, axis1=0, axis2=1)

1.2. Scalars 91
NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
a : array_like
Array from which the diagonals are taken.
offset : int, optional
Offset of the diagonal from the main diagonal. Can be positive or negative. Defaults to
main diagonal (0).
axis1 : int, optional
Axis to be used as the first axis of the 2-D sub-arrays from which the diagonals should
be taken. Defaults to first axis (0).
axis2 : int, optional
Axis to be used as the second axis of the 2-D sub-arrays from which the diagonals
should be taken. Defaults to second axis (1).
Returns
array_of_diagonals : ndarray
If a is 2-D, a 1-D array containing the diagonal is returned. If the dimension of a is
larger, then an array of diagonals is returned, “packed” from left-most dimension to
right-most (e.g., if a is 3-D, then the diagonals are “packed” along rows).
Raises
ValueError :
If the dimension of a is less than 2.
See Also:

diag
MATLAB work-a-like for 1-D and 2-D arrays.
diagflat
Create diagonal arrays.
trace
Sum along diagonals.

Examples
>>> a = np.arange(4).reshape(2,2)
>>> a
array([[0, 1],
[2, 3]])
>>> a.diagonal()
array([0, 3])
>>> a.diagonal(1)
array([1])

92 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

... 1) # the "middle" (row) axis first.

array([[0, 6],
[1, 7]])

numpy.mean(a, axis=None, dtype=None, out=None)

average
Weighted average

1.2. Scalars 93
NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.mean(a)
2.5
>>> np.mean(a, axis=0)
array([ 2., 3.])
>>> np.mean(a, axis=1)
array([ 1.5, 3.5])

In single precision, mean can be inaccurate:

>>> a = np.zeros((2, 512*512), dtype=np.float32)
>>> a[0, :] = 1.0
>>> a[1, :] = 0.1
>>> np.mean(a)
0.546875

Computing the mean in float64 is more accurate:

>>> np.mean(a, dtype=np.float64)
0.55000000074505806

To group the indices by element, rather than dimension, use:

transpose(nonzero(a))

The result of this is always a 2-D array, with a row for each non-zero element.
Parameters
a : array_like
Input array.
Returns
tuple_of_arrays : tuple
Indices of elements that are non-zero.
See Also:

Examples

94 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x = np.eye(3)
>>> x
array([[ 1., 0., 0.],
[ 0., 1., 0.],
[ 0., 0., 1.]])
>>> np.nonzero(x)
(array([0, 1, 2]), array([0, 1, 2]))

>>> x[np.nonzero(x)]
array([ 1., 1., 1.])
>>> np.transpose(np.nonzero(x))
array([[0, 0],
[1, 1],
[2, 2]])

The nonzero method of the boolean array can also be called.

>>> (a > 3).nonzero()
(array([1, 1, 1, 2, 2, 2]), array([0, 1, 2, 0, 1, 2]))

numpy.prod(a, axis=None, dtype=None, out=None)

1.2. Scalars 95
NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:

ndarray.prod
equivalent method
numpy.doc.ufuncs
Section “Output arguments”

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow. That means that, on a
32-bit platform:
>>> x = np.array([536870910, 536870910, 536870910, 536870910])
>>> np.prod(x) #random
16

Examples
By default, calculate the product of all elements:
>>> np.prod([1.,2.])
2.0

Even when the input array is two-dimensional:

>>> np.prod([[1.,2.],[3.,4.]])
24.0

But we can also specify the axis over which to multiply:

>>> np.prod([[1.,2.],[3.,4.]], axis=1)
array([ 2., 12.])

If the type of x is unsigned, then the output type is the unsigned platform integer:
>>> x = np.array([1, 2, 3], dtype=np.uint8)
>>> np.prod(x).dtype == np.uint
True

If x is of a signed integer type, then the output type is the default platform integer:
>>> x = np.array([1, 2, 3], dtype=np.int8)
>>> np.prod(x).dtype == np.int
True

numpy.ptp(a, axis=None, out=None)

96 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Alternative output array in which to place the result. It must have the same shape and
buffer length as the expected output, but the type of the output values will be cast if
necessary.
Returns
ptp : ndarray
A new array holding the result, unless out was specified, in which case a reference to
out is returned.

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.ptp(x, axis=0)

array([2, 2])

>>> np.ptp(x, axis=1)

array([1, 1])

numpy.put(a, ind, v, mode=’raise’)

Replaces specified elements of an array with given values.
The indexing works on the flattened target array. put is roughly equivalent to:
a.flat[ind] = v

See Also:
putmask, place

1.2. Scalars 97
NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> a = np.arange(5)
>>> np.put(a, [0, 2], [-44, -55])
>>> a
array([-44, 1, -55, 3, 4])

>>> a = np.arange(5)
>>> np.put(a, 22, -5, mode=’clip’)
>>> a
array([ 0, 1, 2, 3, -5])

ndarray.flat
1-D iterator over an array.
ndarray.flatten
1-D array copy of the elements of an array in row-major order.

Examples
It is equivalent to reshape(-1, order=order).
>>> x = np.array([[1, 2, 3], [4, 5, 6]])
>>> print np.ravel(x)
[1 2 3 4 5 6]

>>> print x.reshape(-1)

[1 2 3 4 5 6]

98 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> print np.ravel(x, order=’F’)

[1 4 2 5 3 6]

When order is ‘A’, it will preserve the array’s ‘C’ or ‘F’ ordering:
>>> print np.ravel(x.T)
[1 4 2 5 3 6]
>>> print np.ravel(x.T, order=’A’)
[1 2 3 4 5 6]

When order is ‘K’, it will preserve orderings that are neither ‘C’ nor ‘F’, but won’t reverse axes:
>>> a = np.arange(3)[::-1]; a
array([2, 1, 0])
>>> a.ravel(order=’C’)
array([2, 1, 0])
>>> a.ravel(order=’K’)
array([2, 1, 0])

numpy.repeat(a, repeats, axis=None)

tile
Tile an array.

Examples
>>> x = np.array([[1,2],[3,4]])
>>> np.repeat(x, 2)

1.2. Scalars 99
NumPy Reference, Release 2.0.0.dev-ac3cba3

array([1, 1, 2, 2, 3, 3, 4, 4])
>>> np.repeat(x, 3, axis=1)
array([[1, 1, 1, 2, 2, 2],
[3, 3, 3, 4, 4, 4]])
>>> np.repeat(x, [1, 2], axis=0)
array([[1, 2],
[3, 4],
[3, 4]])

numpy.reshape(a, newshape, order=’C’)

Gives a new shape to an array without changing its data.
Parameters
a : array_like
Array to be reshaped.
newshape : int or tuple of ints
The new shape should be compatible with the original shape. If an integer, then the
result will be a 1-D array of that length. One shape dimension can be -1. In this case,
the value is inferred from the length of the array and remaining dimensions.
order : {‘C’, ‘F’, ‘A’}, optional
Determines whether the array data should be viewed as in C (row-major) order, FOR-
TRAN (column-major) order, or the C/FORTRAN order should be preserved.
Returns
reshaped_array : ndarray
This will be a new view object if possible; otherwise, it will be a copy.
See Also:

ndarray.reshape
Equivalent method.

Examples
>>> a = np.array([[1,2,3], [4,5,6]])
>>> np.reshape(a, 6)
array([1, 2, 3, 4, 5, 6])
>>> np.reshape(a, 6, order=’F’)
array([1, 4, 2, 5, 3, 6])

100 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.reshape(a, (3,-1)) # the unspecified value is inferred to be 2

array([[1, 2],
[3, 4],
[5, 6]])

numpy.resize(a, new_shape)
Return a new array with the specified shape.
If the new array is larger than the original array, then the new array is filled with repeated copies of a. Note
that this behavior is different from a.resize(new_shape) which fills with zeros instead of repeated copies of
a.
Parameters
a : array_like
Array to be resized.
new_shape : int or tuple of int
Shape of resized array.
Returns
reshaped_array : ndarray
The new array is formed from the data in the old array, repeated if necessary to fill out
the required number of elements. The data are repeated in the order that they are stored
in memory.
See Also:

ndarray.resize
resize an array in-place.

Examples
>>> a=np.array([[0,1],[2,3]])
>>> np.resize(a,(1,4))
array([[0, 1, 2, 3]])
>>> np.resize(a,(2,4))
array([[0, 1, 2, 3],
[0, 1, 2, 3]])

1.2. Scalars 101

NumPy Reference, Release 2.0.0.dev-ac3cba3

Returns
indices : array of ints
Array of insertion points with the same shape as v.
See Also:

sort
Return a sorted copy of an array.
histogram
Produce histogram from 1-D data.

Notes
Binary search is used to find the required insertion points.
As of Numpy 1.4.0 searchsorted works with real/complex arrays containing nan values. The enhanced
sort order is documented in sort.

Examples
>>> np.searchsorted([1,2,3,4,5], 3)
2
>>> np.searchsorted([1,2,3,4,5], 3, side=’right’)
3
>>> np.searchsorted([1,2,3,4,5], [-10, 10, 2, 3])
array([0, 5, 1, 2])

numpy.sort(a, axis=-1, kind=’quicksort’, order=None)

ndarray.sort
Method to sort an array in-place.
argsort
Indirect sort.

102 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

lexsort
Indirect stable sort on multiple keys.
searchsorted
Find elements in a sorted array.

Notes
The various sorting algorithms are characterized by their average speed, worst case performance, work
space size, and whether they are stable. A stable sort keeps items with the same key in the same relative
order. The three available algorithms have the following properties:
kind speed worst case work space stable
‘quicksort’ 1 O(n^2) 0 no
‘mergesort’ 2 O(n*log(n)) ~n/2 yes
‘heapsort’ 3 O(n*log(n)) 0 no
All the sort algorithms make temporary copies of the data when sorting along any but the last axis. Con-
sequently, sorting along the last axis is faster and uses less space than sorting along any other axis.
The sort order for complex numbers is lexicographic. If both the real and imaginary parts are non-nan then
the order is determined by the real parts except when they are equal, in which case the order is determined
by the imaginary parts.
Previous to numpy 1.4.0 sorting real and complex arrays containing nan values led to undefined behaviour.
In numpy versions >= 1.4.0 nan values are sorted to the end. The extended sort order is:
•Real: [R, nan]
•Complex: [R + Rj, R + nanj, nan + Rj, nan + nanj]
where R is a non-nan real value. Complex values with the same nan placements are sorted according to
the non-nan part if it exists. Non-nan values are sorted as before.

Sort by age, then height if ages are equal:

>>> np.sort(a, order=[’age’, ’height’])
array([(’Galahad’, 1.7, 38), (’Lancelot’, 1.8999999999999999, 38),

1.2. Scalars 103

NumPy Reference, Release 2.0.0.dev-ac3cba3

(’Arthur’, 1.8, 41)],

dtype=[(’name’, ’|S10’), (’height’, ’<f8’), (’age’, ’<i4’)])

numpy.squeeze(a)
Remove single-dimensional entries from the shape of an array.
Parameters
a : array_like
Input data.
Returns
squeezed : ndarray
The input array, but with with all dimensions of length 1 removed. Whenever possible,
a view on a is returned.

Examples
>>> x = np.array([[[0], [1], [2]]])
>>> x.shape
(1, 3, 1)
>>> np.squeeze(x).shape
(3,)

numpy.std(a, axis=None, dtype=None, out=None, ddof=0)

104 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:
var, mean
numpy.doc.ufuncs
Section “Output arguments”

Notes
The standard deviation is the square root of the average of the squared deviations from the mean, i.e., std
= sqrt(mean(abs(x - x.mean())**2)).
The average squared deviation is normally calculated as x.sum() / N, where N = len(x). If, how-
ever, ddof is specified, the divisor N - ddof is used instead. In standard statistical practice, ddof=1
provides an unbiased estimator of the variance of the infinite population. ddof=0 provides a maximum
likelihood estimate of the variance for normally distributed variables. The standard deviation computed in
this function is the square root of the estimated variance, so even with ddof=1, it will not be an unbiased
estimate of the standard deviation per se.
Note that, for complex numbers, std takes the absolute value before squaring, so that the result is always
real and nonnegative.
For floating-point input, the std is computed using the same precision the input has. Depending on the input
data, this can cause the results to be inaccurate, especially for float32 (see example below). Specifying a
higher-accuracy accumulator using the dtype keyword can alleviate this issue.

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.std(a)
1.1180339887498949
>>> np.std(a, axis=0)
array([ 1., 1.])
>>> np.std(a, axis=1)
array([ 0.5, 0.5])

In single precision, std() can be inaccurate:

>>> a = np.zeros((2,512*512), dtype=np.float32)
>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.std(a)
0.45172946707416706

Computing the standard deviation in float64 is more accurate:

>>> np.std(a, dtype=np.float64)
0.44999999925552653

numpy.sum(a, axis=None, dtype=None, out=None)

1.2. Scalars 105

NumPy Reference, Release 2.0.0.dev-ac3cba3

The type of the returned array and of the accumulator in which the elements are
summed. By default, the dtype of a is used. An exception is when a has an integer
type with less precision than the default platform integer. In that case, the default plat-
form integer is used instead.
out : ndarray, optional
Array into which the output is placed. By default, a new array is created. If out is
given, it must be of the appropriate shape (the shape of a with axis removed, i.e.,
numpy.delete(a.shape, axis)). Its type is preserved. See doc.ufuncs (Sec-
tion “Output arguments”) for more details.
Returns
sum_along_axis : ndarray
An array with the same shape as a, with the specified axis removed. If a is a 0-d array,
or if axis is None, a scalar is returned. If an output array is specified, a reference to out
is returned.
See Also:

ndarray.sum
Equivalent method.
cumsum
Cumulative sum of array elements.
trapz
Integration of array values using the composite trapezoidal rule.

mean, average

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

If the accumulator is too small, overflow occurs:

>>> np.ones(128, dtype=np.int8).sum(dtype=np.int8)
-128

numpy.swapaxes(a, axis1, axis2)

Interchange two axes of an array.
Parameters
a : array_like
Input array.

106 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

axis1 : int
First axis.
axis2 : int
Second axis.
Returns
a_swapped : ndarray
If a is an ndarray, then a view of a is returned; otherwise a new array is created.

Examples
>>> x = np.array([[1,2,3]])
>>> np.swapaxes(x,0,1)
array([[1],
[2],
[3]])

>>> x = np.array([[[0,1],[2,3]],[[4,5],[6,7]]])
>>> x
array([[[0, 1],
[2, 3]],
[[4, 5],
[6, 7]]])

>>> np.swapaxes(x,0,2)
array([[[0, 4],
[2, 6]],
[[1, 5],
[3, 7]]])

numpy.take(a, indices, axis=None, out=None, mode=’raise’)

1.2. Scalars 107

NumPy Reference, Release 2.0.0.dev-ac3cba3

• ‘clip’ – clip to the range

‘clip’ mode means that all indices that are too large are replaced by the index that ad-
dresses the last element along that axis. Note that this disables indexing with negative
numbers.
Returns
subarray : ndarray
The returned array has the same type as a.
See Also:

ndarray.take
equivalent method

Examples
>>> a = [4, 3, 5, 7, 6, 8]
>>> indices = [0, 1, 4]
>>> np.take(a, indices)
array([4, 3, 6])

In this example if a is an ndarray, “fancy” indexing can be used.

>>> a = np.array(a)
>>> a[indices]
array([4, 3, 6])

numpy.trace(a, offset=0, axis1=0, axis2=1, dtype=None, out=None)

108 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Array into which the output is placed. Its type is preserved and it must be of the right
shape to hold the output.
Returns
sum_along_diagonals : ndarray
If a is 2-D, the sum along the diagonal is returned. If a has larger dimensions, then an
array of sums along diagonals is returned.
See Also:
diag, diagonal, diagflat

Examples
>>> np.trace(np.eye(3))
3.0
>>> a = np.arange(8).reshape((2,2,2))
>>> np.trace(a)
array([6, 8])

>>> a = np.arange(24).reshape((2,2,2,3))
>>> np.trace(a).shape
(2, 3)

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.transpose(x)
array([[0, 2],
[1, 3]])

>>> x = np.ones((1, 2, 3))

>>> np.transpose(x, (1, 0, 2)).shape
(2, 1, 3)

1.2. Scalars 109

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.var(a, axis=None, dtype=None, out=None, ddof=0)

Compute the variance along the specified axis.
Returns the variance of the array elements, a measure of the spread of a distribution. The variance is
computed for the flattened array by default, otherwise over the specified axis.
Parameters
a : array_like
Array containing numbers whose variance is desired. If a is not an array, a conversion
is attempted.
axis : int, optional
Axis along which the variance is computed. The default is to compute the variance of
the flattened array.
dtype : data-type, optional
Type to use in computing the variance. For arrays of integer type the default is float32;
for arrays of float types it is the same as the array type.
out : ndarray, optional
Alternate output array in which to place the result. It must have the same shape as the
expected output, but the type is cast if necessary.
ddof : int, optional
“Delta Degrees of Freedom”: the divisor used in the calculation is N - ddof, where
N represents the number of elements. By default ddof is zero.
Returns
variance : ndarray, see dtype parameter above
If out=None, returns a new array containing the variance; otherwise, a reference to
the output array is returned.
See Also:

std
Standard deviation
mean
Average
numpy.doc.ufuncs
Section “Output arguments”

110 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

For floating-point input, the variance is computed using the same precision the input has. Depending
on the input data, this can cause the results to be inaccurate, especially for float32 (see example below).
Specifying a higher-accuracy accumulator using the dtype keyword can alleviate this issue.

Examples
>>> a = np.array([[1,2],[3,4]])
>>> np.var(a)
1.25
>>> np.var(a,0)
array([ 1., 1.])
>>> np.var(a,1)
array([ 0.25, 0.25])

In single precision, var() can be inaccurate:

>>> a = np.zeros((2,512*512), dtype=np.float32)
>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.var(a)
0.20405951142311096

Computing the standard deviation in float64 is more accurate:

>>> np.var(a, dtype=np.float64)
0.20249999932997387
>>> ((1-0.55)**2 + (0.1-0.55)**2)/2
0.20250000000000001

generic.__array__()
sc.__array__(|type) return 0-dim array
generic.__array_wrap__()
sc.__array_wrap__(obj) return scalar from array
generic.squeeze()
Not implemented (virtual attribute)
Class generic exists solely to derive numpy scalars from, and possesses, albeit unimplemented, all the attributes
of the ndarray class so as to provide a uniform API.
See Also:
The
generic.byteswap()
Not implemented (virtual attribute)
Class generic exists solely to derive numpy scalars from, and possesses, albeit unimplemented, all the attributes
of the ndarray class so as to provide a uniform API.
See Also:
The
generic.__reduce__()

generic.__setstate__()

generic.setflags()
Not implemented (virtual attribute)

1.2. Scalars 111

NumPy Reference, Release 2.0.0.dev-ac3cba3

Class generic exists solely to derive numpy scalars from, and possesses, albeit unimplemented, all the attributes
of the ndarray class so as to provide a uniform API.
See Also:
The

1.2.5 Defining new types

There are two ways to effectively define a new array scalar type (apart from composing record dtypes from the built-in
scalar types): One way is to simply subclass the ndarray and overwrite the methods of interest. This will work to a
degree, but internally certain behaviors are fixed by the data type of the array. To fully customize the data type of an
array you need to define a new data-type, and register it with NumPy. Such new types can only be defined in C, using
the Numpy C-API.

1.3 Data type objects (dtype)

A data type object (an instance of numpy.dtype class) describes how the bytes in the fixed-size block of memory
corresponding to an array item should be interpreted. It describes the following aspects of the data:
1. Type of the data (integer, float, Python object, etc.)
2. Size of the data (how many bytes is in e.g. the integer)
3. Byte order of the data (little-endian or big-endian)
4. If the data type is a record, an aggregate of other data types, (e.g., describing an array item consisting of an
integer and a float),
(a) what are the names of the “fields” of the record, by which they can be accessed,
(b) what is the data-type of each field, and
(c) which part of the memory block each field takes.
5. If the data is a sub-array, what is its shape and data type.
To describe the type of scalar data, there are several built-in scalar types in Numpy for various precision of integers,
floating-point numbers, etc. An item extracted from an array, e.g., by indexing, will be a Python object whose type is
the scalar type associated with the data type of the array.
Note that the scalar types are not dtype objects, even though they can be used in place of one whenever a data type
specification is needed in Numpy. Record data types are formed by creating a data type whose fields contain other
data types. Each field has a name by which it can be accessed. The parent data type should be of sufficient size to
contain all its fields; the parent can for example be based on the void type which allows an arbitrary item size. Record
data types may also contain other record types and fixed-size sub-array data types in their fields. Finally, a data type
can describe items that are themselves arrays of items of another data type. These sub-arrays must, however, be of a
fixed size. If an array is created using a data-type describing a sub-array, the dimensions of the sub-array are appended
to the shape of the array when the array is created. Sub-arrays in a field of a record behave differently, see Record
Access.
Example
A simple data type containing a 32-bit big-endian integer: (see Specifying and constructing data types for details on
construction)

112 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> dt = np.dtype(’>i4’)
>>> dt.byteorder
’>’
>>> dt.itemsize
4
>>> dt.name
’int32’
>>> dt.type is np.int32
True

The corresponding array scalar type is int32.

Example
A record data type containing a 16-character string (in field ‘name’) and a sub-array of two 64-bit floating-point
number (in field ‘grades’):
>>> dt = np.dtype([(’name’, np.str_, 16), (’grades’, np.float64, (2,))])
>>> dt[’name’]
dtype(’|S16’)
>>> dt[’grades’]
dtype((’float64’,(2,)))

Items of an array of this data type are wrapped in an array scalar type that also has two fields:
>>> x = np.array([(’Sarah’, (8.0, 7.0)), (’John’, (6.0, 7.0))], dtype=dt)
>>> x[1]
(’John’, [6.0, 7.0])
>>> x[1][’grades’]
array([ 6., 7.])
>>> type(x[1])
<type ’numpy.void’>
>>> type(x[1][’grades’])
<type ’numpy.ndarray’>

1.3.1 Specifying and constructing data types

Whenever a data-type is required in a NumPy function or method, either a dtype object or something that can be
converted to one can be supplied. Such conversions are done by the dtype constructor:
dtype Create a data type object.
class numpy.dtype
Create a data type object.
A numpy array is homogeneous, and contains elements described by a dtype object. A dtype object can be
constructed from different combinations of fundamental numeric types.
Parameters
obj :
Object to be converted to a data type object.
align : bool, optional
Add padding to the fields to match what a C compiler would output for a similar C-
struct. Can be True only if obj is a dictionary or a comma-separated string.
copy : bool, optional

1.3. Data type objects (dtype) 113

NumPy Reference, Release 2.0.0.dev-ac3cba3

Make a new copy of the data-type object. If False, the result may just be a reference
to a built-in data-type object.

Examples
Using array-scalar type:
>>> np.dtype(np.int16)
dtype(’int16’)

Record, one field name ‘f1’, containing int16:

>>> np.dtype([(’f1’, np.int16)])
dtype([(’f1’, ’<i2’)])

Record, one field named ‘f1’, in itself containing a record with one field:
>>> np.dtype([(’f1’, [(’f1’, np.int16)])])
dtype([(’f1’, [(’f1’, ’<i2’)])])

Record, two fields: the first field contains an unsigned int, the second an int32:
>>> np.dtype([(’f1’, np.uint), (’f2’, np.int32)])
dtype([(’f1’, ’<u4’), (’f2’, ’<i4’)])

Using array-protocol type strings:

>>> np.dtype([(’a’,’f8’),(’b’,’S10’)])
dtype([(’a’, ’<f8’), (’b’, ’|S10’)])

Using comma-separated field formats. The shape is (2,3):

>>> np.dtype("i4, (2,3)f8")
dtype([(’f0’, ’<i4’), (’f1’, ’<f8’, (2, 3))])

Using dictionaries. Two fields named ‘gender’ and ‘age’:

>>> np.dtype({’names’:[’gender’,’age’], ’formats’:[’S1’,np.uint8]})
dtype([(’gender’, ’|S1’), (’age’, ’|u1’)])

Offsets in bytes, here 0 and 25:

>>> np.dtype({’surname’:(’S25’,0),’age’:(np.uint8,25)})
dtype([(’surname’, ’|S25’), (’age’, ’|u1’)])

Methods
newbyteorder
What can be converted to a data-type object is described below:
dtype object

114 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Used as-is.
None
The default data type: float_.
Array-scalar types
The 24 built-in array scalar type objects all convert to an associated data-type object. This is true for their
sub-classes as well.
Note that not all data-type information can be supplied with a type-object: for example, flexible data-types
have a default itemsize of 0, and require an explicitly given size to be useful.
Example
>>> dt = np.dtype(np.int32) # 32-bit integer
>>> dt = np.dtype(np.complex128) # 128-bit complex floating-point number

Generic types
The generic hierarchical type objects convert to corresponding type objects according to the associations:
number, inexact, floating float
complexfloating cfloat
integer, signedinteger int_
unsignedinteger uint
character string
generic, flexible void
Built-in Python types
Several python types are equivalent to a corresponding array scalar when used to generate a dtype object:
int int_
bool bool_
float float_
complex cfloat
str string
unicode unicode_
buffer void
(all others) object_
Example
>>> dt = np.dtype(float) # Python-compatible floating-point number
>>> dt = np.dtype(int) # Python-compatible integer
>>> dt = np.dtype(object) # Python object

Types with .dtype

Any type object with a dtype attribute: The attribute will be accessed and used directly. The attribute
must return something that is convertible into a dtype object.
Several kinds of strings can be converted. Recognized strings can be prepended with ’>’ (big-endian), ’<’ (little-
endian), or ’=’ (hardware-native, the default), to specify the byte order.
One-character strings
Each built-in data-type has a character code (the updated Numeric typecodes), that uniquely identifies it.
Example

1.3. Data type objects (dtype) 115

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> dt = np.dtype(’b’) # byte, native byte order

>>> dt = np.dtype(’>H’) # big-endian unsigned short
>>> dt = np.dtype(’<f’) # little-endian single-precision float
>>> dt = np.dtype(’d’) # double-precision floating-point number

Array-protocol type strings (see The Array Interface)

The first character specifies the kind of data and the remaining characters specify how many bytes of data.
The supported kinds are
’b’ Boolean
’i’ (signed) integer
’u’ unsigned integer
’f’ floating-point
’c’ complex-floating point
’S’, ’a’ string
’U’ unicode
’V’ anything (void)
Example
>>> dt = np.dtype(’i4’) # 32-bit signed integer
>>> dt = np.dtype(’f8’) # 64-bit floating-point number
>>> dt = np.dtype(’c16’) # 128-bit complex floating-point number
>>> dt = np.dtype(’a25’) # 25-character string

String with comma-separated fields

Numarray introduced a short-hand notation for specifying the format of a record as a comma-separated
string of basic formats.
A basic format in this context is an optional shape specifier followed by an array-protocol type string.
Parenthesis are required on the shape if it is greater than 1-d. NumPy allows a modification on the format
in that any string that can uniquely identify the type can be used to specify the data-type in a field. The
generated data-type fields are named ’f0’, ’f1’, ..., ’f<N-1>’ where N (>1) is the number of comma-
separated basic formats in the string. If the optional shape specifier is provided, then the data-type for the
corresponding field describes a sub-array.
Example
• field named f0 containing a 32-bit integer
• field named f1 containing a 2 x 3 sub-array of 64-bit floating-point numbers
• field named f2 containing a 32-bit floating-point number
>>> dt = np.dtype("i4, (2,3)f8, f4")

• field named f0 containing a 3-character string

• field named f1 containing a sub-array of shape (3,) containing 64-bit unsigned integers
• field named f2 containing a 3 x 4 sub-array containing 10-character strings
>>> dt = np.dtype("a3, 3u8, (3,4)a10")

Type strings
Any string in numpy.sctypeDict.keys():
Example

116 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> dt = np.dtype(’uint32’) # 32-bit unsigned integer

>>> dt = np.dtype(’Float64’) # 64-bit floating-point number

(flexible_dtype, itemsize)
The first argument must be an object that is converted to a flexible data-type object (one whose element
size is 0), the second argument is an integer providing the desired itemsize.
Example
>>> dt = np.dtype((void, 10)) # 10-byte wide data block
>>> dt = np.dtype((str, 35)) # 35-character string
>>> dt = np.dtype((’U’, 10)) # 10-character unicode string

(fixed_dtype, shape)
The first argument is any object that can be converted into a fixed-size data-type object. The second
argument is the desired shape of this type. If the shape parameter is 1, then the data-type object is
equivalent to fixed dtype. If shape is a tuple, then the new dtype defines a sub-array of the given shape.
Example
>>> dt = np.dtype((np.int32, (2,2))) # 2 x 2 integer sub-array
>>> dt = np.dtype((’S10’, 1)) # 10-character string
>>> dt = np.dtype((’i4, (2,3)f8, f4’, (2,3))) # 2 x 3 record sub-array

(base_dtype, new_dtype)
Both arguments must be convertible to data-type objects in this case. The base_dtype is the data-type
object that the new data-type builds on. This is how you could assign named fields to any built-in data-
type object.
Example
32-bit integer, whose first two bytes are interpreted as an integer via field real, and the following two
bytes via field imag.
>>> dt = np.dtype((np.int32,{’real’:(np.int16, 0),’imag’:(np.int16, 2)})

32-bit integer, which is interpreted as consisting of a sub-array of shape (4,) containing 8-bit integers:
>>> dt = np.dtype((np.int32, (np.int8, 4)))

32-bit integer, containing fields r, g, b, a that interpret the 4 bytes in the integer as four unsigned integers:
>>> dt = np.dtype((’i4’, [(’r’,’u1’),(’g’,’u1’),(’b’,’u1’),(’a’,’u1’)]))

[(field_name, field_dtype, field_shape), ...]

obj should be a list of fields where each field is described by a tuple of length 2 or 3. (Equivalent to the
descr item in the __array_interface__ attribute.)
The first element, field_name, is the field name (if this is ” then a standard field name, ’f#’, is assigned).
The field name may also be a 2-tuple of strings where the first string is either a “title” (which may be any
string or unicode string) or meta-data for the field which can be any object, and the second string is the
“name” which must be a valid Python identifier.
The second element, field_dtype, can be anything that can be interpreted as a data-type.
The optional third element field_shape contains the shape if this field represents an array of the data-type
in the second element. Note that a 3-tuple with a third argument equal to 1 is equivalent to a 2-tuple.

1.3. Data type objects (dtype) 117

NumPy Reference, Release 2.0.0.dev-ac3cba3

This style does not accept align in the dtype constructor as it is assumed that all of the memory is
accounted for by the array interface description.
Example
Data-type with fields big (big-endian 32-bit integer) and little (little-endian 32-bit integer):
>>> dt = np.dtype([(’big’, ’>i4’), (’little’, ’<i4’)])

Data-type with fields R, G, B, A, each being an unsigned 8-bit integer:

>>> dt = np.dtype([(’R’,’u1’), (’G’,’u1’), (’B’,’u1’), (’A’,’u1’)])

{’names’: ..., ’formats’: ..., ’offsets’: ..., ’titles’: ...}

This style has two required and two optional keys. The names and formats keys are required. Their
respective values are equal-length lists with the field names and the field formats. The field names must
be strings and the field formats can be any object accepted by dtype constructor.
The optional keys in the dictionary are offsets and titles and their values must each be lists of the same
length as the names and formats lists. The offsets value is a list of byte offsets (integers) for each field,
while the titles value is a list of titles for each field (None can be used if no title is desired for that field).
The titles can be any string or unicode object and will add another entry to the fields dictionary
keyed by the title and referencing the same field tuple which will contain the title as an additional tuple
member.
Example
Data type with fields r, g, b, a, each being a 8-bit unsigned integer:
>>> dt = np.dtype({’names’: [’r’,’g’,’b’,’a’],
... ’formats’: [uint8, uint8, uint8, uint8]})

Data type with fields r and b (with the given titles), both being 8-bit unsigned integers, the first at byte
position 0 from the start of the field and the second at position 2:
>>> dt = np.dtype({’names’: [’r’,’b’], ’formats’: [’u1’, ’u1’],
... ’offsets’: [0, 2],
... ’titles’: [’Red pixel’, ’Blue pixel’]})

{’field1’: ..., ’field2’: ..., ...}

This style allows passing in the fields attribute of a data-type object.
obj should contain string or unicode keys that refer to (data-type, offset) or (data-type,
offset, title) tuples.
Example
Data type containing field col1 (10-character string at byte position 0), col2 (32-bit float at byte posi-
tion 10), and col3 (integers at byte position 14):
>>> dt = np.dtype({’col1’: (’S10’, 0), ’col2’: (float32, 10),
’col3’: (int, 14)})

1.3.2 dtype

Numpy data type descriptions are instances of the dtype class.

118 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Attributes

The type of the data is described by the following dtype attributes:

dtype.type The type object used to instantiate a scalar of this data-type.
dtype.kind A character code (one of ‘biufcSUV’) identifying the general kind of data.
dtype.char A unique character code for each of the 21 different built-in types.
dtype.num A unique number for each of the 21 different built-in types.
dtype.str The array-protocol typestring of this data-type object.
dtype.type
The type object used to instantiate a scalar of this data-type.
dtype.kind
A character code (one of ‘biufcSUV’) identifying the general kind of data.
dtype.char
A unique character code for each of the 21 different built-in types.
dtype.num
A unique number for each of the 21 different built-in types.
These are roughly ordered from least-to-most precision.
dtype.str
The array-protocol typestring of this data-type object.
Size of the data is in turn described by:
dtype.name A bit-width name for this data-type.
dtype.itemsize The element size of this data-type object.
dtype.name
A bit-width name for this data-type.
Un-sized flexible data-type objects do not have this attribute.
dtype.itemsize
The element size of this data-type object.
For 18 of the 21 types this number is fixed by the data-type. For the flexible data-types, this number can be
anything.
Endianness of this data:
dtype.byteorder A character indicating the byte-order of this data-type object.
dtype.byteorder
A character indicating the byte-order of this data-type object.
One of:
‘=’ native
‘<’ little-endian
‘>’ big-endian
‘|’ not applicable
All built-in data-type objects have byteorder either ‘=’ or ‘|’.

Examples
>>> dt = np.dtype(’i2’)
>>> dt.byteorder
’=’

1.3. Data type objects (dtype) 119

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> # endian is not relevant for 8 bit numbers

>>> np.dtype(’i1’).byteorder
’|’
>>> # or ASCII strings
>>> np.dtype(’S2’).byteorder
’|’
>>> # Even if specific code is given, and it is native
>>> # ’=’ is the byteorder
>>> import sys
>>> sys_is_le = sys.byteorder == ’little’
>>> native_code = sys_is_le and ’<’ or ’>’
>>> swapped_code = sys_is_le and ’>’ or ’<’
>>> dt = np.dtype(native_code + ’i2’)
>>> dt.byteorder
’=’
>>> # Swapped code shows up as itself
>>> dt = np.dtype(swapped_code + ’i2’)
>>> dt.byteorder == swapped_code
True

Information about sub-data-types in a record:

dtype.fields Dictionary of named fields defined for this data type, or None.
dtype.names Ordered list of field names, or None if there are no fields.
dtype.fields
Dictionary of named fields defined for this data type, or None.
The dictionary is indexed by keys that are the names of the fields. Each entry in the dictionary is a tuple fully
describing the field:
(dtype, offset[, title])

If present, the optional title can be any object (if it is a string or unicode then it will also be a key in the fields
dictionary, otherwise it’s meta-data). Notice also that the first two elements of the tuple can be passed directly
as arguments to the ndarray.getfield and ndarray.setfield methods.
See Also:
ndarray.getfield, ndarray.setfield

Examples
>>> dt = np.dtype([(’name’, np.str_, 16), (’grades’, np.float64, (2,))])
>>> print dt.fields
{’grades’: (dtype((’float64’,(2,))), 16), ’name’: (dtype(’|S16’), 0)}

dtype.names
Ordered list of field names, or None if there are no fields.
The names are ordered according to increasing byte offset. This can be used, for example, to walk through all
of the named fields in offset order.

Examples
>>> dt = np.dtype([(’name’, np.str_, 16), (’grades’, np.float64, (2,))])
>>> dt.names
(’name’, ’grades’)

For data types that describe sub-arrays:

120 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

dtype.subdtype Tuple (item_dtype, shape) if this dtype describes a sub-array, and

dtype.shape Shape tuple of the sub-array if this data type describes a sub-array,
dtype.subdtype
Tuple (item_dtype, shape) if this dtype describes a sub-array, and None otherwise.
The shape is the fixed shape of the sub-array described by this data type, and item_dtype the data type of the
array.
If a field whose dtype object has this attribute is retrieved, then the extra dimensions implied by shape are tacked
on to the end of the retrieved array.
dtype.shape
Shape tuple of the sub-array if this data type describes a sub-array, and () otherwise.
Attributes providing additional information:
dtype.hasobjectBoolean indicating whether this dtype contains any reference-counted objects in any fields
or sub-dtypes.
dtype.flags Bit-flags describing how this data type is to be interpreted.
dtype.isbuiltinInteger indicating how this dtype relates to the built-in dtypes.
dtype.isnative Boolean indicating whether the byte order of this dtype is native
dtype.descr Array-interface compliant full description of the data-type.
dtype.alignmentThe required alignment (bytes) of this data-type according to the compiler.
dtype.hasobject
Boolean indicating whether this dtype contains any reference-counted objects in any fields or sub-dtypes.
Recall that what is actually in the ndarray memory representing the Python object is the memory address of that
object (a pointer). Special handling may be required, and this attribute is useful for distinguishing data types
that may contain arbitrary Python objects and data-types that won’t.
dtype.flags
Bit-flags describing how this data type is to be interpreted.
Bit-masks are in numpy.core.multiarray as the constants ITEM_HASOBJECT, LIST_PICKLE,
ITEM_IS_POINTER, NEEDS_INIT, NEEDS_PYAPI, USE_GETITEM, USE_SETITEM. A full explanation of
these flags is in C-API documentation; they are largely useful for user-defined data-types.
dtype.isbuiltin
Integer indicating how this dtype relates to the built-in dtypes.
Read-only.
0 if this is a structured array type, with fields
1 if this is a dtype compiled into numpy (such as ints, floats etc)
2 if the dtype is for a user-defined numpy type A user-defined type uses the numpy C-API machinery to
extend numpy to handle a new array type. See user.user-defined-data-types in the Numpy manual.

Examples
>>> dt = np.dtype(’i2’)
>>> dt.isbuiltin
1
>>> dt = np.dtype(’f8’)
>>> dt.isbuiltin
1
>>> dt = np.dtype([(’field1’, ’f8’)])
>>> dt.isbuiltin
0

1.3. Data type objects (dtype) 121

NumPy Reference, Release 2.0.0.dev-ac3cba3

dtype.isnative
Boolean indicating whether the byte order of this dtype is native to the platform.
dtype.descr
Array-interface compliant full description of the data-type.
The format is that required by the ‘descr’ key in the __array_interface__ attribute.
dtype.alignment
The required alignment (bytes) of this data-type according to the compiler.
More information is available in the C-API section of the manual.

Methods

Data types have the following method for changing the byte order:
dtype.newbyteorder([new_order]) Return a new dtype with a different byte order.
dtype.newbyteorder(new_order=’S’)
Return a new dtype with a different byte order.
Changes are also made in all fields and sub-arrays of the data type.
Parameters
new_order : string, optional
Byte order to force; a value from the byte order specifications below. The default value
(‘S’) results in swapping the current byte order. new_order codes can be any of:

* ’S’ - swap dtype from current to opposite endian

* {’<’, ’L’} - little endian
* {’>’, ’B’} - big endian
* {’=’, ’N’} - native order
* {’|’, ’I’} - ignore (no change to byte order)

The code does a case-insensitive check on the first letter of new_order for these alter-
natives. For example, any of ‘>’ or ‘B’ or ‘b’ or ‘brian’ are valid to specify big-endian.
Returns
new_dtype : dtype
New dtype object with the given change to the byte order.

Notes
Changes are also made in all fields and sub-arrays of the data type.

Examples
>>> import sys
>>> sys_is_le = sys.byteorder == ’little’
>>> native_code = sys_is_le and ’<’ or ’>’
>>> swapped_code = sys_is_le and ’>’ or ’<’
>>> native_dt = np.dtype(native_code+’i2’)
>>> swapped_dt = np.dtype(swapped_code+’i2’)
>>> native_dt.newbyteorder(’S’) == swapped_dt
True
>>> native_dt.newbyteorder() == swapped_dt
True
>>> native_dt == swapped_dt.newbyteorder(’S’)

122 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

True
>>> native_dt == swapped_dt.newbyteorder(’=’)
True
>>> native_dt == swapped_dt.newbyteorder(’N’)
True
>>> native_dt == native_dt.newbyteorder(’|’)
True
>>> np.dtype(’<i2’) == native_dt.newbyteorder(’<’)
True
>>> np.dtype(’<i2’) == native_dt.newbyteorder(’L’)
True
>>> np.dtype(’>i2’) == native_dt.newbyteorder(’>’)
True
>>> np.dtype(’>i2’) == native_dt.newbyteorder(’B’)
True

The following methods implement the pickle protocol:

dtype.__reduce__
dtype.__setstate__
dtype.__reduce__()

dtype.__setstate__()

1.4 Indexing

ndarrays can be indexed using the standard Python x[obj] syntax, where x is the array and obj the selection.
There are three kinds of indexing available: record access, basic slicing, advanced indexing. Which one occurs
depends on obj.
Note: In Python, x[(exp1, exp2, ..., expN)] is equivalent to x[exp1, exp2, ..., expN]; the
latter is just syntactic sugar for the former.

1.4.1 Basic Slicing

Basic slicing extends Python’s basic concept of slicing to N dimensions. Basic slicing occurs when obj is a slice
object (constructed by start:stop:step notation inside of brackets), an integer, or a tuple of slice objects and
integers. Ellipsis and newaxis objects can be interspersed with these as well. In order to remain backward
compatible with a common usage in Numeric, basic slicing is also initiated if the selection object is any sequence
(such as a list) containing slice objects, the Ellipsis object, or the newaxis object, but no integer arrays or
other embedded sequences. The simplest case of indexing with N integers returns an array scalar representing the
corresponding item. As in Python, all indices are zero-based: for the i-th index ni , the valid range is 0 ≤ ni < di
where di is the i-th element of the shape of the array. Negative indices are interpreted as counting from the end of the
array (i.e., if i < 0, it means ni + i).
All arrays generated by basic slicing are always views of the original array.
The standard rules of sequence slicing apply to basic slicing on a per-dimension basis (including using a step index).
Some useful concepts to remember include:
• The basic slice syntax is i:j:k where i is the starting index, j is the stopping index, and k is the step (k 6= 0).
This selects the m elements (in the corresponding dimension) with index values i, i + k, ..., i + (m - 1) k where

1.4. Indexing 123

NumPy Reference, Release 2.0.0.dev-ac3cba3

m = q + (r 6= 0) and q and r are the quotient and remainder obtained by dividing j - i by k: j - i = q k + r, so

that i + (m - 1) k < j.
Example
>>> x = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> x[1:7:2]
array([1, 3, 5])

• Negative i and j are interpreted as n + i and n + j where n is the number of elements in the corresponding
dimension. Negative k makes stepping go towards smaller indices.
Example
>>> x[-2:10]
array([8, 9])
>>> x[-3:3:-1]
array([7, 6, 5, 4])

• Assume n is the number of elements in the dimension being sliced. Then, if i is not given it defaults to 0 for k >
0 and n for k < 0 . If j is not given it defaults to n for k > 0 and -1 for k < 0 . If k is not given it defaults to 1.
Note that :: is the same as : and means select all indices along this axis.
Example
>>> x[5:]
array([5, 6, 7, 8, 9])

• If the number of objects in the selection tuple is less than N , then : is assumed for any subsequent dimensions.
Example
>>> x = np.array([[[1],[2],[3]], [[4],[5],[6]]])
>>> x.shape
(2, 3, 1)
>>> x[1:2]
array([[[4],
[5],
[6]]])

• Ellipsis expand to the number of : objects needed to make a selection tuple of the same length as x.ndim.
Only the first ellipsis is expanded, any others are interpreted as :.
Example
>>> x[...,0]
array([[1, 2, 3],
[4, 5, 6]])

• Each newaxis object in the selection tuple serves to expand the dimensions of the resulting selection by one
unit-length dimension. The added dimension is the position of the newaxis object in the selection tuple.
Example
>>> x[:,np.newaxis,:,:].shape
(2, 1, 3, 1)

• An integer, i, returns the same values as i:i+1 except the dimensionality of the returned object is reduced by 1.
In particular, a selection tuple with the p-th element an integer (and all other entries :) returns the corresponding
sub-array with dimension N - 1. If N = 1 then the returned object is an array scalar. These objects are explained
in Scalars.

124 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

• If the selection tuple has all entries : except the p-th entry which is a slice object i:j:k, then the returned
array has dimension N formed by concatenating the sub-arrays returned by integer indexing of elements i, i+k,
..., i + (m - 1) k < j,
• Basic slicing with more than one non-: entry in the slicing tuple, acts like repeated application of slicing using
a single non-: entry, where the non-: entries are successively taken (with all other non-: entries replaced by
:). Thus, x[ind1,...,ind2,:] acts like x[ind1][...,ind2,:] under basic slicing.

Warning: The above is not true for advanced slicing.

• You may use slicing to set values in the array, but (unlike lists) you can never grow the array. The size of the
value to be set in x[obj] = value must be (broadcastable) to the same shape as x[obj].
Note: Remember that a slicing tuple can always be constructed as obj and used in the x[obj] notation. Slice objects
can be used in the construction in place of the [start:stop:step] notation. For example, x[1:10:5,::-1]
can also be implemented as obj = (slice(1,10,5), slice(None,None,-1)); x[obj] . This can be
useful for constructing generic code that works on arrays of arbitrary dimension.
numpy.newaxis
The newaxis object can be used in the basic slicing syntax discussed above. None can also be used instead
of newaxis.

1.4.2 Advanced indexing

Advanced indexing is triggered when the selection object, obj, is a non-tuple sequence object, an ndarray (of data
type integer or bool), or a tuple with at least one sequence object or ndarray (of data type integer or bool). There are
two types of advanced indexing: integer and Boolean.
Advanced indexing always returns a copy of the data (contrast with basic slicing that returns a view).

Integer

Integer indexing allows selection of arbitrary items in the array based on their N-dimensional index. This kind of
selection occurs when advanced indexing is triggered and the selection object is not an array of data type bool. For the
discussion below, when the selection object is not a tuple, it will be referred to as if it had been promoted to a 1-tuple,
which will be called the selection tuple. The rules of advanced integer-style indexing are:
• If the length of the selection tuple is larger than N an error is raised.
• All sequences and scalars in the selection tuple are converted to intp indexing arrays.
• All selection tuple objects must be convertible to intp arrays, slice objects, or the Ellipsis object.
• The first Ellipsis object will be expanded, and any other Ellipsis objects will be treated as full slice (:)
objects. The expanded Ellipsis object is replaced with as many full slice (:) objects as needed to make the
length of the selection tuple N .
• If the selection tuple is smaller than N, then as many : objects as needed are added to the end of the selection
tuple so that the modified selection tuple has length N.
• All the integer indexing arrays must be broadcastable to the same shape.
• The shape of the output (or the needed shape of the object to be used for setting) is the broadcasted shape.
• After expanding any ellipses and filling out any missing : objects in the selection tuple, then let Nt be the
number of indexing arrays, and let Ns = N − Nt be the number of slice objects. Note that Nt > 0 (or we
wouldn’t be doing advanced integer indexing).

1.4. Indexing 125

NumPy Reference, Release 2.0.0.dev-ac3cba3

• If Ns = 0 then the M-dimensional result is constructed by varying the index tuple (i_1, ..., i_M) over
the range of the result shape and for each value of the index tuple (ind_1, ..., ind_M):
result[i_1, ..., i_M] == x[ind_1[i_1, ..., i_M], ind_2[i_1, ..., i_M],
..., ind_N[i_1, ..., i_M]]

Example
Suppose the shape of the broadcasted indexing arrays is 3-dimensional and N is 2. Then the result is found by
letting i, j, k run over the shape found by broadcasting ind_1 and ind_2, and each i, j, k yields:
result[i,j,k] = x[ind_1[i,j,k], ind_2[i,j,k]]

• If Ns > 0, then partial indexing is done. This can be somewhat mind-boggling to understand, but if you think
in terms of the shapes of the arrays involved, it can be easier to grasp what happens. In simple cases (i.e.
one indexing array and N - 1 slice objects) it does exactly what you would expect (concatenation of repeated
application of basic slicing). The rule for partial indexing is that the shape of the result (or the interpreted shape
of the object to be used in setting) is the shape of x with the indexed subspace replaced with the broadcasted
indexing subspace. If the index subspaces are right next to each other, then the broadcasted indexing space
directly replaces all of the indexed subspaces in x. If the indexing subspaces are separated (by slice objects),
then the broadcasted indexing space is first, followed by the sliced subspace of x.
Example
Suppose x.shape is (10,20,30) and ind is a (2,3,4)-shaped indexing intp array, then result =
x[...,ind,:] has shape (10,2,3,4,30) because the (20,)-shaped subspace has been replaced with a
(2,3,4)-shaped broadcasted indexing subspace. If we let i, j, k loop over the (2,3,4)-shaped subspace then
result[...,i,j,k,:] = x[...,ind[i,j,k],:]. This example produces the same result as
x.take(ind, axis=-2).
Example
Now let x.shape be (10,20,30,40,50) and suppose ind_1 and ind_2 are broadcastable to the shape (2,3,4).
Then x[:,ind_1,ind_2] has shape (10,2,3,4,40,50) because the (20,30)-shaped subspace from X has
been replaced with the (2,3,4) subspace from the indices. However, x[:,ind_1,:,ind_2] has shape
(2,3,4,10,30,50) because there is no unambiguous place to drop in the indexing subspace, thus it is tacked-
on to the beginning. It is always possible to use .transpose() to move the subspace anywhere desired.
(Note that this example cannot be replicated using take.)

Boolean

This advanced indexing occurs when obj is an array object of Boolean type (such as may be returned from com-
parison operators). It is always equivalent to (but faster than) x[obj.nonzero()] where, as described above,
obj.nonzero() returns a tuple (of length obj.ndim) of integer index arrays showing the True elements of obj.
The special case when obj.ndim == x.ndim is worth mentioning. In this case x[obj] returns a 1-dimensional
array filled with the elements of x corresponding to the True values of obj. The search order will be C-style (last
index varies the fastest). If obj has True values at entries that are outside of the bounds of x, then an index error will
be raised.
You can also use Boolean arrays as element of the selection tuple. In such instances, they will always be interpreted
as nonzero(obj) and the equivalent integer indexing will be done.

Warning: The definition of advanced indexing means that x[(1,2,3),] is fundamentally different than
x[(1,2,3)]. The latter is equivalent to x[1,2,3] which will trigger basic selection while the former will
trigger advanced indexing. Be sure to understand why this is occurs.
Also recognize that x[[1,2,3]] will trigger advanced indexing, whereas x[[1,2,slice(None)]] will
trigger basic slicing.

126 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

1.4.3 Record Access

See Also:
Data type objects (dtype), Scalars
If the ndarray object is a record array, i.e. its data type is a record data type, the fields of the array can be accessed
by indexing the array with strings, dictionary-like.
Indexing x[’field-name’] returns a new view to the array, which is of the same shape as x (except when the field
is a sub-array) but of data type x.dtype[’field-name’] and contains only the part of the data in the specified
field. Also record array scalars can be “indexed” this way.
If the accessed field is a sub-array, the dimensions of the sub-array are appended to the shape of the result.
Example
>>> x = np.zeros((2,2), dtype=[(’a’, np.int32), (’b’, np.float64, (3,3))])
>>> x[’a’].shape
(2, 2)
>>> x[’a’].dtype
dtype(’int32’)
>>> x[’b’].shape
(2, 2, 3, 3)
>>> x[’b’].dtype
dtype(’float64’)

1.4.4 Flat Iterator indexing

x.flat returns an iterator that will iterate over the entire array (in C-contiguous style with the last index varying
the fastest). This iterator object can also be indexed using basic slicing or advanced indexing as long as the selection
object is not a tuple. This should be clear from the fact that x.flat is a 1-dimensional view. It can be used for integer
indexing with 1-dimensional C-style-flat indices. The shape of any returned array is therefore the shape of the integer
indexing object.

1.5 Standard array subclasses

The ndarray in NumPy is a “new-style” Python built-in-type. Therefore, it can be inherited from (in Python or in
C) if desired. Therefore, it can form a foundation for many useful classes. Often whether to sub-class the array object
or to simply use the core array component as an internal part of a new class is a difficult decision, and can be simply a
matter of choice. NumPy has several tools for simplifying how your new object interacts with other array objects, and
so the choice may not be significant in the end. One way to simplify the question is by asking yourself if the object
you are interested in can be replaced as a single array or does it really require two or more arrays at its core.
Note that asarray always returns the base-class ndarray. If you are confident that your use of the array object can
handle any subclass of an ndarray, then asanyarray can be used to allow subclasses to propagate more cleanly
through your subroutine. In principal a subclass could redefine any aspect of the array and therefore, under strict
guidelines, asanyarray would rarely be useful. However, most subclasses of the arrayobject will not redefine
certain aspects of the array object such as the buffer interface, or the attributes of the array. One important example,
however, of why your subroutine may not be able to handle an arbitrary subclass of an array is that matrices redefine
the “*” operator to be matrix-multiplication, rather than element-by-element multiplication.

1.5. Standard array subclasses 127

NumPy Reference, Release 2.0.0.dev-ac3cba3

1.5.1 Special attributes and methods

See Also:
Subclassing ndarray
Numpy provides several hooks that subclasses of ndarray can customize:
numpy.__array_finalize__(self )
This method is called whenever the system internally allocates a new array from obj, where obj is a subclass
(subtype) of the ndarray. It can be used to change attributes of self after construction (so as to ensure
a 2-d matrix for example), or to update meta-information from the “parent.” Subclasses inherit a default
implementation of this method that does nothing.
numpy.__array_prepare__(array, context=None)
At the beginning of every ufunc, this method is called on the input object with the highest array priority, or the
output object if one was specified. The output array is passed in and whatever is returned is passed to the ufunc.
Subclasses inherit a default implementation of this method which simply returns the output array unmodified.
Subclasses may opt to use this method to transform the output array into an instance of the subclass and update
metadata before returning the array to the ufunc for computation.
numpy.__array_wrap__(array, context=None)
At the end of every ufunc, this method is called on the input object with the highest array priority, or the output
object if one was specified. The ufunc-computed array is passed in and whatever is returned is passed to the
user. Subclasses inherit a default implementation of this method, which transforms the array into a new instance
of the object’s class. Subclasses may opt to use this method to transform the output array into an instance of the
subclass and update metadata before returning the array to the user.
numpy.__array_priority__
The value of this attribute is used to determine what type of object to return in situations where there is more
than one possibility for the Python type of the returned object. Subclasses inherit a default value of 1.0 for this
attribute.
numpy.__array__([dtype ])
If a class having the __array__ method is used as the output object of an ufunc, results will be written to the
object returned by __array__.

1.5.2 Matrix objects

matrix objects inherit from the ndarray and therefore, they have the same attributes and methods of ndarrays. There
are six important differences of matrix objects, however, that may lead to unexpected results when you use matrices
but expect them to act like arrays:
1. Matrix objects can be created using a string notation to allow Matlab-style syntax where spaces separate columns
and semicolons (‘;’) separate rows.
2. Matrix objects are always two-dimensional. This has far-reaching implications, in that m.ravel() is still two-
dimensional (with a 1 in the first dimension) and item selection returns two-dimensional objects so that sequence
behavior is fundamentally different than arrays.
3. Matrix objects over-ride multiplication to be matrix-multiplication. Make sure you understand this for func-
tions that you may want to receive matrices. Especially in light of the fact that asanyarray(m) returns a
matrix when m is a matrix.
4. Matrix objects over-ride power to be matrix raised to a power. The same warning about using power inside a
function that uses asanyarray(...) to get an array object holds for this fact.
5. The default __array_priority__ of matrix objects is 10.0, and therefore mixed operations with ndarrays always
produce matrices.

128 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

6. Matrices have special attributes which make calculations easier. These are
matrix.T transpose
matrix.H hermitian (conjugate) transpose
matrix.I inverse
matrix.A base array
matrix.T
transpose
matrix.H
hermitian (conjugate) transpose
matrix.I
inverse
matrix.A
base array

Warning: Matrix objects over-ride multiplication, ‘*’, and power, ‘**’, to be matrix-multiplication and matrix
power, respectively. If your subroutine can accept sub-classes and you do not convert to baseclass arrays, then
you must use the ufuncs multiply and power to be sure that you are performing the correct operation for all inputs.

The matrix class is a Python subclass of the ndarray and can be used as a reference for how to construct your own
subclass of the ndarray. Matrices can be created from other matrices, strings, and anything else that can be converted
to an ndarray . The name “mat “is an alias for “matrix “in NumPy.
matrix Returns a matrix from an array-like object, or from a string of data.
asmatrix(data[, dtype]) Interpret the input as a matrix.
bmat(obj[, ldict, gdict]) Build a matrix object from a string, nested sequence, or array.
class numpy.matrix
Returns a matrix from an array-like object, or from a string of data. A matrix is a specialized 2-D array that
retains its 2-D nature through operations. It has certain special operators, such as * (matrix multiplication) and
** (matrix power).
Parameters
data : array_like or string
If data is a string, it is interpreted as a matrix with commas or spaces separating
columns, and semicolons separating rows.
dtype : data-type
Data-type of the output matrix.
copy : bool
If data is already an ndarray, then this flag determines whether the data is copied (the
default), or whether a view is constructed.
See Also:
array

Examples
>>> a = np.matrix(’1 2; 3 4’)
>>> print a
[[1 2]
[3 4]]

1.5. Standard array subclasses 129

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.matrix([[1, 2], [3, 4]])

matrix([[1, 2],
[3, 4]])

Methods

all(a[, axis, out]) Test whether all array elements along a given axis evaluate to True.
any(a[, axis, out]) Test whether any array element along a given axis evaluates to True.
argmax(a[, axis]) Indices of the maximum values along an axis.
argmin(a[, axis]) Return the indices of the minimum values along an axis.
argsort(a[, axis, kind, order]) Returns the indices that would sort an array.
astype
byteswap
choose(a, choices[, out, mode]) Construct an array from an index array and a set of arrays to choose from.
clip(a, a_min, a_max[, out]) Clip (limit) the values in an array.
compress(condition, a[, axis, out]) Return selected slices of an array along given axis.
conj() Return the complex conjugate, element-wise.
conjugate() Return the complex conjugate, element-wise.
copy(a) Return an array copy of the given object.
cumprod(a[, axis, dtype, out]) Return the cumulative product of elements along a given axis.
cumsum(a[, axis, dtype, out]) Return the cumulative sum of the elements along a given axis.
diagonal(a[, offset, axis1, axis2]) Return specified diagonals.
dot(a, b[, out]) Dot product of two arrays.
dump
dumps
fill
flatten
getA
getA1
getH
getI
getT
getfield
item
itemset
max(a[, axis, out]) Return the maximum of an array or maximum along an axis.
mean(a[, axis, dtype, out]) Compute the arithmetic mean along the specified axis.
min(a[, axis, out]) Return the minimum of an array or minimum along an axis.
newbyteorder
nonzero(a) Return the indices of the elements that are non-zero.
prod(a[, axis, dtype, out]) Return the product of array elements over a given axis.
ptp(a[, axis, out]) Range of values (maximum - minimum) along an axis.
put(a, ind, v[, mode]) Replaces specified elements of an array with given values.
ravel(a[, order]) Return a flattened array.
repeat(a, repeats[, axis]) Repeat elements of an array.
reshape(a, newshape[, order]) Gives a new shape to an array without changing its data.
resize(a, new_shape) Return a new array with the specified shape.
round(a[, decimals, out]) Round an array to the given number of decimals.
searchsorted(a, v[, side]) Find indices where elements should be inserted to maintain order.
setasflat
setfield
setflags
sort(a[, axis, kind, order]) Return a sorted copy of an array.
Continued on next page

130 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Table 1.3 – continued from previous page

squeeze(a) Remove single-dimensional entries from the shape of an array.
std(a[, axis, dtype, out, ddof]) Compute the standard deviation along the specified axis.
sum(a[, axis, dtype, out]) Sum of array elements over a given axis.
swapaxes(a, axis1, axis2) Interchange two axes of an array.
take(a, indices[, axis, out, mode]) Take elements from an array along an axis.
tofile
tolist
tostring
trace(a[, offset, axis1, axis2, dtype, out]) Return the sum along diagonals of the array.
transpose(a[, axes]) Permute the dimensions of an array.
var(a[, axis, dtype, out, ddof]) Compute the variance along the specified axis.
view

numpy.all(a, axis=None, out=None)

ndarray.all
equivalent method
any
Test whether any element along a given axis evaluates to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.all([[True,False],[True,True]])
False

1.5. Standard array subclasses 131

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.all([[True,False],[True,True]], axis=0)

array([ True, False], dtype=bool)

>>> np.all([-1, 4, 5])

True

>>> np.all([1.0, np.nan])

True

>>> o=np.array([False])
>>> z=np.all([-1, 4, 5], out=o)
>>> id(z), id(o), z
(28293632, 28293632, array([ True], dtype=bool))

numpy.any(a, axis=None, out=None)

ndarray.any
equivalent method
all
Test whether all elements along a given axis evaluate to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.any([[True, False], [True, True]])
True

132 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.any([[True, False], [False, False]], axis=0)

array([ True, False], dtype=bool)

>>> np.any([-1, 0, 5])

True

>>> np.any(np.nan)
True

Notes
In case of multiple occurrences of the maximum values, the indices corresponding to the first occurrence
are returned.

Examples
>>> a = np.arange(6).reshape(2,3)
>>> a
array([[0, 1, 2],
[3, 4, 5]])
>>> np.argmax(a)
5
>>> np.argmax(a, axis=0)
array([1, 1, 1])

1.5. Standard array subclasses 133

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.argmax(a, axis=1)

array([2, 2])

>>> b = np.arange(6)
>>> b[1] = 5
>>> b
array([0, 5, 2, 3, 4, 5])
>>> np.argmax(b) # Only the first occurrence is returned.
1

numpy.argmin(a, axis=None)
Return the indices of the minimum values along an axis.
See Also:

argmax
Similar function. Please refer to numpy.argmax for detailed documentation.

numpy.argsort(a, axis=-1, kind=’quicksort’, order=None)

sort
Describes sorting algorithms used.
lexsort
Indirect stable sort with multiple keys.
ndarray.sort
Inplace sort.

134 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Notes
See sort for notes on the different sorting algorithms.
As of NumPy 1.4.0 argsort works with real/complex arrays containing nan values. The enhanced sort
order is documented in sort.

Examples
One dimensional array:
>>> x = np.array([3, 1, 2])
>>> np.argsort(x)
array([1, 2, 0])

Two-dimensional array:
>>> x = np.array([[0, 3], [2, 2]])
>>> x
array([[0, 3],
[2, 2]])

>>> np.argsort(x, axis=0)

array([[0, 1],
[1, 0]])

>>> np.argsort(x, axis=1)

array([[0, 1],
[0, 1]])

Sorting with keys:

>>> x = np.array([(1, 0), (0, 1)], dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])
>>> x
array([(1, 0), (0, 1)],
dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])

>>> np.argsort(x, order=(’x’,’y’))

array([1, 0])

>>> np.argsort(x, order=(’y’,’x’))

array([0, 1])

numpy.choose(a, choices, out=None, mode=’raise’)

1.5. Standard array subclasses 135

NumPy Reference, Release 2.0.0.dev-ac3cba3

•if mode=wrap, values in a (and thus Ba) may be any (signed) integer; modular arithmetic is used to
map integers outside the range [0, n-1] back into that range; and then the new array is constructed as
above;
•if mode=clip, values in a (and thus Ba) may be any (signed) integer; negative integers are mapped
to 0; values greater than n-1 are mapped to n-1; and then the new array is constructed as above.

See Also:

ndarray.choose
equivalent method

136 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

... # 31, etc.

... )
array([20, 31, 12, 3])
>>> np.choose([2, 4, 1, 0], choices, mode=’clip’) # 4 goes to 3 (4-1)
array([20, 31, 12, 3])
>>> # because there are 4 choice arrays
>>> np.choose([2, 4, 1, 0], choices, mode=’wrap’) # 4 goes to (4 mod 4)
array([20, 1, 12, 3])
>>> # i.e., 0

A couple examples illustrating how choose broadcasts:

>>> a = [[1, 0, 1], [0, 1, 0], [1, 0, 1]]
>>> choices = [-10, 10]
>>> np.choose(a, choices)
array([[ 10, -10, 10],
[-10, 10, -10],
[ 10, -10, 10]])

>>> # With thanks to Anne Archibald

numpy.clip(a, a_min, a_max, out=None)

1.5. Standard array subclasses 137

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:

numpy.doc.ufuncs
Section “Output arguments”

numpy.compress(condition, a, axis=None, out=None)

Examples

138 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a = np.array([[1, 2], [3, 4], [5, 6]])

>>> a
array([[1, 2],
[3, 4],
[5, 6]])
>>> np.compress([0, 1], a, axis=0)
array([[3, 4]])
>>> np.compress([False, True, True], a, axis=0)
array([[3, 4],
[5, 6]])
>>> np.compress([False, True], a, axis=1)
array([[2],
[4],
[6]])

Working on the flattened array does not return slices along an axis but selects elements.
>>> np.compress([False, True], a)
array([2])

Examples
>>> np.conjugate(1+2j)
(1-2j)

>>> x = np.eye(2) + 1j * np.eye(2)

>>> np.conjugate(x)
array([[ 1.-1.j, 0.-0.j],
[ 0.-0.j, 1.-1.j]])

numpy.copy(a)
Return an array copy of the given object.
Parameters
a : array_like
Input data.
Returns
arr : ndarray
Array interpretation of a.

Notes
This is equivalent to

1.5. Standard array subclasses 139

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.array(a, copy=True)

Examples
Create an array x, with a reference y and a copy z:
>>> x = np.array([1, 2, 3])
>>> y = x
>>> z = np.copy(x)

Note that, when we modify x, y changes, but not z:

>>> x[0] = 10
>>> x[0] == y[0]
True
>>> x[0] == z[0]
False

numpy.cumprod(a, axis=None, dtype=None, out=None)

numpy.doc.ufuncs
Section “Output arguments”

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

Examples

140 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a = np.array([1,2,3])
>>> np.cumprod(a) # intermediate results 1, 1*2
... # total product 1*2*3 = 6
array([1, 2, 6])
>>> a = np.array([[1, 2, 3], [4, 5, 6]])
>>> np.cumprod(a, dtype=float) # specify type of output
array([ 1., 2., 6., 24., 120., 720.])

The cumulative product for each column (i.e., over the rows) of a:
>>> np.cumprod(a, axis=0)
array([[ 1, 2, 3],
[ 4, 10, 18]])

The cumulative product for each row (i.e. over the columns) of a:
>>> np.cumprod(a,axis=1)
array([[ 1, 2, 6],
[ 4, 20, 120]])

numpy.cumsum(a, axis=None, dtype=None, out=None)

sum
Sum array elements.
trapz
Integration of array values using the composite trapezoidal rule.

1.5. Standard array subclasses 141

NumPy Reference, Release 2.0.0.dev-ac3cba3

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

>>> np.cumsum(a,axis=0) # sum over rows for each of the 3 columns

array([[1, 2, 3],
[5, 7, 9]])
>>> np.cumsum(a,axis=1) # sum over columns for each of the 2 rows
array([[ 1, 3, 6],
[ 4, 9, 15]])

numpy.diagonal(a, offset=0, axis1=0, axis2=1)

142 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:

diag
MATLAB work-a-like for 1-D and 2-D arrays.
diagflat
Create diagonal arrays.
trace
Sum along diagonals.

Examples
>>> a = np.arange(4).reshape(2,2)
>>> a
array([[0, 1],
[2, 3]])
>>> a.diagonal()
array([0, 3])
>>> a.diagonal(1)
array([1])

Parameters
a : array_like
First argument.
b : array_like

1.5. Standard array subclasses 143

NumPy Reference, Release 2.0.0.dev-ac3cba3

Second argument.
out : ndarray, optional
Output argument. This must have the exact kind that would be returned if it was not
used. In particular, it must have the right type, must be C-contiguous, and its dtype
must be the dtype that would be returned for dot(a,b). This is a performance feature.
Therefore, if these conditions are not met, an exception is raised, instead of attempting
to be flexible.
Returns
output : ndarray
Returns the dot product of a and b. If a and b are both scalars or both 1-D arrays then a
scalar is returned; otherwise an array is returned. If out is given, then it is returned.
Raises
ValueError :
If the last dimension of a is not the same size as the second-to-last dimension of b.

See Also:

vdot
Complex-conjugating dot product.
tensordot
Sum products over arbitrary axes.
einsum
Einstein summation convention.

Examples
>>> np.dot(3, 4)
12

Neither argument is complex-conjugated:

>>> np.dot([2j, 3j], [2j, 3j])
(-13+0j)

For 2-D arrays it’s the matrix product:

>>> a = [[1, 0], [0, 1]]
>>> b = [[4, 1], [2, 2]]
>>> np.dot(a, b)
array([[4, 1],
[2, 2]])

>>> a = np.arange(3*4*5*6).reshape((3,4,5,6))
>>> b = np.arange(3*4*5*6)[::-1].reshape((5,4,6,3))
>>> np.dot(a, b)[2,3,2,1,2,2]
499128
>>> sum(a[2,3,2,:] * b[1,2,:,2])
499128

numpy.mean(a, axis=None, dtype=None, out=None)

144 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
a : array_like
Array containing numbers whose mean is desired. If a is not an array, a conversion is
attempted.
axis : int, optional
Axis along which the means are computed. The default is to compute the mean of the
flattened array.
dtype : data-type, optional
Type to use in computing the mean. For integer inputs, the default is float64; for floating
point inputs, it is the same as the input dtype.
out : ndarray, optional
Alternate output array in which to place the result. The default is None; if provided, it
must have the same shape as the expected output, but the type will be cast if necessary.
See doc.ufuncs for details.
Returns
m : ndarray, see dtype parameter above
If out=None, returns a new array containing the mean values, otherwise a reference to
the output array is returned.
See Also:

average
Weighted average

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.mean(a)
2.5
>>> np.mean(a, axis=0)
array([ 2., 3.])
>>> np.mean(a, axis=1)
array([ 1.5, 3.5])

In single precision, mean can be inaccurate:

>>> a = np.zeros((2, 512*512), dtype=np.float32)
>>> a[0, :] = 1.0
>>> a[1, :] = 0.1
>>> np.mean(a)
0.546875

Computing the mean in float64 is more accurate:

1.5. Standard array subclasses 145

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.mean(a, dtype=np.float64)

0.55000000074505806

To group the indices by element, rather than dimension, use:

transpose(nonzero(a))

The result of this is always a 2-D array, with a row for each non-zero element.
Parameters
a : array_like
Input array.
Returns
tuple_of_arrays : tuple
Indices of elements that are non-zero.
See Also:

Examples
>>> x = np.eye(3)
>>> x
array([[ 1., 0., 0.],
[ 0., 1., 0.],
[ 0., 0., 1.]])
>>> np.nonzero(x)
(array([0, 1, 2]), array([0, 1, 2]))

>>> x[np.nonzero(x)]
array([ 1., 1., 1.])
>>> np.transpose(np.nonzero(x))
array([[0, 0],
[1, 1],
[2, 2]])

146 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a = np.array([[1,2,3],[4,5,6],[7,8,9]])
>>> a > 3
array([[False, False, False],
[ True, True, True],
[ True, True, True]], dtype=bool)
>>> np.nonzero(a > 3)
(array([1, 1, 1, 2, 2, 2]), array([0, 1, 2, 0, 1, 2]))

The nonzero method of the boolean array can also be called.

>>> (a > 3).nonzero()
(array([1, 1, 1, 2, 2, 2]), array([0, 1, 2, 0, 1, 2]))

numpy.prod(a, axis=None, dtype=None, out=None)

ndarray.prod
equivalent method
numpy.doc.ufuncs
Section “Output arguments”

1.5. Standard array subclasses 147

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
By default, calculate the product of all elements:
>>> np.prod([1.,2.])
2.0

Even when the input array is two-dimensional:

>>> np.prod([[1.,2.],[3.,4.]])
24.0

But we can also specify the axis over which to multiply:

>>> np.prod([[1.,2.],[3.,4.]], axis=1)
array([ 2., 12.])

If the type of x is unsigned, then the output type is the unsigned platform integer:
>>> x = np.array([1, 2, 3], dtype=np.uint8)
>>> np.prod(x).dtype == np.uint
True

If x is of a signed integer type, then the output type is the default platform integer:
>>> x = np.array([1, 2, 3], dtype=np.int8)
>>> np.prod(x).dtype == np.int
True

numpy.ptp(a, axis=None, out=None)

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.ptp(x, axis=0)

array([2, 2])

148 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.ptp(x, axis=1)

array([1, 1])

numpy.put(a, ind, v, mode=’raise’)

Replaces specified elements of an array with given values.
The indexing works on the flattened target array. put is roughly equivalent to:
a.flat[ind] = v

See Also:
putmask, place

Examples
>>> a = np.arange(5)
>>> np.put(a, [0, 2], [-44, -55])
>>> a
array([-44, 1, -55, 3, 4])

>>> a = np.arange(5)
>>> np.put(a, 22, -5, mode=’clip’)
>>> a
array([ 0, 1, 2, 3, -5])

numpy.ravel(a, order=’C’)
Return a flattened array.
A 1-D array, containing the elements of the input, is returned. A copy is made only if needed.
Parameters
a : array_like

1.5. Standard array subclasses 149

NumPy Reference, Release 2.0.0.dev-ac3cba3

Input array. The elements in a are read in the order specified by order, and packed as a
1-D array.
order : {‘C’,’F’, ‘A’, ‘K’}, optional
The elements of a are read in this order. ‘C’ means to view the elements in C (row-
major) order. ‘F’ means to view the elements in Fortran (column-major) order. ‘A’
means to view the elements in ‘F’ order if a is Fortran contiguous, ‘C’ order otherwise.
‘K’ means to view the elements in the order they occur in memory, except for reversing
the data when strides are negative. By default, ‘C’ order is used.
Returns
1d_array : ndarray
Output of the same dtype as a, and of shape (a.size(),).
See Also:

ndarray.flat
1-D iterator over an array.
ndarray.flatten
1-D array copy of the elements of an array in row-major order.

Examples
It is equivalent to reshape(-1, order=order).
>>> x = np.array([[1, 2, 3], [4, 5, 6]])
>>> print np.ravel(x)
[1 2 3 4 5 6]

>>> print x.reshape(-1)

[1 2 3 4 5 6]

>>> print np.ravel(x, order=’F’)

[1 4 2 5 3 6]

When order is ‘A’, it will preserve the array’s ‘C’ or ‘F’ ordering:
>>> print np.ravel(x.T)
[1 4 2 5 3 6]
>>> print np.ravel(x.T, order=’A’)
[1 2 3 4 5 6]

150 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.repeat(a, repeats, axis=None)

tile
Tile an array.

numpy.reshape(a, newshape, order=’C’)

Gives a new shape to an array without changing its data.
Parameters
a : array_like
Array to be reshaped.
newshape : int or tuple of ints

1.5. Standard array subclasses 151

NumPy Reference, Release 2.0.0.dev-ac3cba3

The new shape should be compatible with the original shape. If an integer, then the
result will be a 1-D array of that length. One shape dimension can be -1. In this case,
the value is inferred from the length of the array and remaining dimensions.
order : {‘C’, ‘F’, ‘A’}, optional
Determines whether the array data should be viewed as in C (row-major) order, FOR-
TRAN (column-major) order, or the C/FORTRAN order should be preserved.
Returns
reshaped_array : ndarray
This will be a new view object if possible; otherwise, it will be a copy.
See Also:

ndarray.reshape
Equivalent method.

Examples
>>> a = np.array([[1,2,3], [4,5,6]])
>>> np.reshape(a, 6)
array([1, 2, 3, 4, 5, 6])
>>> np.reshape(a, 6, order=’F’)
array([1, 4, 2, 5, 3, 6])

>>> np.reshape(a, (3,-1)) # the unspecified value is inferred to be 2

array([[1, 2],
[3, 4],
[5, 6]])

152 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Returns
reshaped_array : ndarray
The new array is formed from the data in the old array, repeated if necessary to fill out
the required number of elements. The data are repeated in the order that they are stored
in memory.
See Also:

ndarray.resize
resize an array in-place.

Examples
>>> a=np.array([[0,1],[2,3]])
>>> np.resize(a,(1,4))
array([[0, 1, 2, 3]])
>>> np.resize(a,(2,4))
array([[0, 1, 2, 3],
[0, 1, 2, 3]])

sort
Return a sorted copy of an array.
histogram
Produce histogram from 1-D data.

Notes
Binary search is used to find the required insertion points.
As of Numpy 1.4.0 searchsorted works with real/complex arrays containing nan values. The enhanced
sort order is documented in sort.

1.5. Standard array subclasses 153

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> np.searchsorted([1,2,3,4,5], 3)
2
>>> np.searchsorted([1,2,3,4,5], 3, side=’right’)
3
>>> np.searchsorted([1,2,3,4,5], [-10, 10, 2, 3])
array([0, 5, 1, 2])

numpy.sort(a, axis=-1, kind=’quicksort’, order=None)

ndarray.sort
Method to sort an array in-place.
argsort
Indirect sort.
lexsort
Indirect stable sort on multiple keys.
searchsorted
Find elements in a sorted array.

154 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

The sort order for complex numbers is lexicographic. If both the real and imaginary parts are non-nan then
the order is determined by the real parts except when they are equal, in which case the order is determined
by the imaginary parts.
Previous to numpy 1.4.0 sorting real and complex arrays containing nan values led to undefined behaviour.
In numpy versions >= 1.4.0 nan values are sorted to the end. The extended sort order is:
•Real: [R, nan]
•Complex: [R + Rj, R + nanj, nan + Rj, nan + nanj]
where R is a non-nan real value. Complex values with the same nan placements are sorted according to
the non-nan part if it exists. Non-nan values are sorted as before.

Sort by age, then height if ages are equal:

1.5. Standard array subclasses 155

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> x = np.array([[[0], [1], [2]]])
>>> x.shape
(1, 3, 1)
>>> np.squeeze(x).shape
(3,)

numpy.std(a, axis=None, dtype=None, out=None, ddof=0)

156 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Note that, for complex numbers, std takes the absolute value before squaring, so that the result is always
real and nonnegative.
For floating-point input, the std is computed using the same precision the input has. Depending on the input
data, this can cause the results to be inaccurate, especially for float32 (see example below). Specifying a
higher-accuracy accumulator using the dtype keyword can alleviate this issue.

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.std(a)
1.1180339887498949
>>> np.std(a, axis=0)
array([ 1., 1.])
>>> np.std(a, axis=1)
array([ 0.5, 0.5])

In single precision, std() can be inaccurate:

>>> a = np.zeros((2,512*512), dtype=np.float32)
>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.std(a)
0.45172946707416706

Computing the standard deviation in float64 is more accurate:

>>> np.std(a, dtype=np.float64)
0.44999999925552653

numpy.sum(a, axis=None, dtype=None, out=None)

1.5. Standard array subclasses 157

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:

ndarray.sum
Equivalent method.
cumsum
Cumulative sum of array elements.
trapz
Integration of array values using the composite trapezoidal rule.

mean, average

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

If the accumulator is too small, overflow occurs:

>>> np.ones(128, dtype=np.int8).sum(dtype=np.int8)
-128

numpy.swapaxes(a, axis1, axis2)

Examples
>>> x = np.array([[1,2,3]])
>>> np.swapaxes(x,0,1)
array([[1],
[2],
[3]])

158 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x = np.array([[[0,1],[2,3]],[[4,5],[6,7]]])
>>> x
array([[[0, 1],
[2, 3]],
[[4, 5],
[6, 7]]])

>>> np.swapaxes(x,0,2)
array([[[0, 4],
[2, 6]],
[[1, 5],
[3, 7]]])

numpy.take(a, indices, axis=None, out=None, mode=’raise’)

ndarray.take
equivalent method

1.5. Standard array subclasses 159

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> a = [4, 3, 5, 7, 6, 8]
>>> indices = [0, 1, 4]
>>> np.take(a, indices)
array([4, 3, 6])

In this example if a is an ndarray, “fancy” indexing can be used.

>>> a = np.array(a)
>>> a[indices]
array([4, 3, 6])

numpy.trace(a, offset=0, axis1=0, axis2=1, dtype=None, out=None)

Examples

160 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.trace(np.eye(3))
3.0
>>> a = np.arange(8).reshape((2,2,2))
>>> np.trace(a)
array([6, 8])

>>> a = np.arange(24).reshape((2,2,2,3))
>>> np.trace(a).shape
(2, 3)

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.transpose(x)
array([[0, 2],
[1, 3]])

>>> x = np.ones((1, 2, 3))

>>> np.transpose(x, (1, 0, 2)).shape
(2, 1, 3)

numpy.var(a, axis=None, dtype=None, out=None, ddof=0)

1.5. Standard array subclasses 161

NumPy Reference, Release 2.0.0.dev-ac3cba3

dtype : data-type, optional

Type to use in computing the variance. For arrays of integer type the default is float32;
for arrays of float types it is the same as the array type.
out : ndarray, optional
Alternate output array in which to place the result. It must have the same shape as the
expected output, but the type is cast if necessary.
ddof : int, optional
“Delta Degrees of Freedom”: the divisor used in the calculation is N - ddof, where
N represents the number of elements. By default ddof is zero.
Returns
variance : ndarray, see dtype parameter above
If out=None, returns a new array containing the variance; otherwise, a reference to
the output array is returned.
See Also:

std
Standard deviation
mean
Average
numpy.doc.ufuncs
Section “Output arguments”

Examples
>>> a = np.array([[1,2],[3,4]])
>>> np.var(a)
1.25
>>> np.var(a,0)
array([ 1., 1.])
>>> np.var(a,1)
array([ 0.25, 0.25])

In single precision, var() can be inaccurate:

162 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a = np.zeros((2,512*512), dtype=np.float32)

>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.var(a)
0.20405951142311096

Computing the standard deviation in float64 is more accurate:

>>> np.var(a, dtype=np.float64)
0.20249999932997387
>>> ((1-0.55)**2 + (0.1-0.55)**2)/2
0.20250000000000001

numpy.asmatrix(data, dtype=None)
Interpret the input as a matrix.
Unlike matrix, asmatrix does not make a copy if the input is already a matrix or an ndarray. Equivalent to
matrix(data, copy=False).
Parameters
data : array_like
Input data.
Returns
mat : matrix
data interpreted as a matrix.

Examples
>>> x = np.array([[1, 2], [3, 4]])

>>> m = np.asmatrix(x)

>>> x[0,0] = 5

>>> m
matrix([[5, 2],
[3, 4]])

numpy.bmat(obj, ldict=None, gdict=None)

Build a matrix object from a string, nested sequence, or array.
Parameters
obj : str or array_like
Input data. Names of variables in the current scope may be referenced, even if obj is a
string.
Returns
out : matrix
Returns a matrix object, which is a specialized 2-D array.
See Also:
matrix

1.5. Standard array subclasses 163

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> A = np.mat(’1 1; 1 1’)
>>> B = np.mat(’2 2; 2 2’)
>>> C = np.mat(’3 4; 5 6’)
>>> D = np.mat(’7 8; 9 0’)

All the following expressions construct the same block matrix:

>>> np.bmat([[A, B], [C, D]])
matrix([[1, 1, 2, 2],
[1, 1, 2, 2],
[3, 4, 7, 8],
[5, 6, 9, 0]])
>>> np.bmat(np.r_[np.c_[A, B], np.c_[C, D]])
matrix([[1, 1, 2, 2],
[1, 1, 2, 2],
[3, 4, 7, 8],
[5, 6, 9, 0]])
>>> np.bmat(’A,B; C,D’)
matrix([[1, 1, 2, 2],
[1, 1, 2, 2],
[3, 4, 7, 8],
[5, 6, 9, 0]])

Example 1: Matrix creation from a string

>>> a=mat(’1 2 3; 4 5 3’)
>>> print (a*a.T).I
[[ 0.2924 -0.1345]
[-0.1345 0.0819]]

Example 2: Matrix creation from nested sequence

>>> mat([[1,5,10],[1.0,3,4j]])
matrix([[ 1.+0.j, 5.+0.j, 10.+0.j],
[ 1.+0.j, 3.+0.j, 0.+4.j]])

Example 3: Matrix creation from an array

>>> mat(random.rand(3,3)).T
matrix([[ 0.7699, 0.7922, 0.3294],
[ 0.2792, 0.0101, 0.9219],
[ 0.3398, 0.7571, 0.8197]])

1.5.3 Memory-mapped file arrays

Memory-mapped files are useful for reading and/or modifying small segments of a large file with regular layout,
without reading the entire file into memory. A simple subclass of the ndarray uses a memory-mapped file for the data
buffer of the array. For small files, the over-head of reading the entire file into memory is typically not significant,
however for large files using memory mapping can save considerable resources.
Memory-mapped-file arrays have one additional method (besides those they inherit from the ndarray): .flush()
which must be called manually by the user to ensure that any changes to the array actually get written to disk.
Note: Memory-mapped arrays use the the Python memory-map object which (prior to Python 2.5) does not allow
files to be larger than a certain size depending on the platform. This size is always < 2GB even on 64-bit systems.
memmap Create a memory-map to an array stored in a binary file on disk.
memmap.flush() Write any changes in the array to the file on disk.

164 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

class numpy.memmap
Create a memory-map to an array stored in a binary file on disk.
Memory-mapped files are used for accessing small segments of large files on disk, without reading the entire
file into memory. Numpy’s memmap’s are array-like objects. This differs from Python’s mmap module, which
uses file-like objects.
Parameters
filename : str or file-like object
The file name or file object to be used as the array data buffer.
dtype : data-type, optional
The data-type used to interpret the file contents. Default is uint8.
mode : {‘r+’, ‘r’, ‘w+’, ‘c’}, optional
The file is opened in this mode:
‘r’ Open existing file for reading only.
‘r+’ Open existing file for reading and writing.
‘w+’ Create or overwrite existing file for reading and writing.
‘c’ Copy-on-write: assignments affect data in memory, but changes are not saved
to disk. The file on disk is read-only.
Default is ‘r+’.
offset : int, optional
In the file, array data starts at this offset. Since offset is measured in bytes, it should be
a multiple of the byte-size of dtype. Requires shape=None. The default is 0.
shape : tuple, optional
The desired shape of the array. By default, the returned array will be 1-D with the
number of elements determined by file size and data-type.
order : {‘C’, ‘F’}, optional
Specify the order of the ndarray memory layout: C (row-major) or Fortran (column-
major). This only has an effect if the shape is greater than 1-D. The default order is
‘C’.

Notes
The memmap object can be used anywhere an ndarray is accepted. Given a memmap fp, isinstance(fp,
numpy.ndarray) returns True.
Memory-mapped arrays use the Python memory-map object which (prior to Python 2.5) does not allow files to
be larger than a certain size depending on the platform. This size is always < 2GB even on 64-bit systems.

Examples
>>> data = np.arange(12, dtype=’float32’)
>>> data.resize((3,4))

This example uses a temporary file so that doctest doesn’t write files to your directory. You would use a ‘normal’
filename.
>>> from tempfile import mkdtemp
>>> import os.path as path
>>> filename = path.join(mkdtemp(), ’newfile.dat’)

1.5. Standard array subclasses 165

NumPy Reference, Release 2.0.0.dev-ac3cba3

Create a memmap with dtype and shape that matches our data:
>>> fp = np.memmap(filename, dtype=’float32’, mode=’w+’, shape=(3,4))
>>> fp
memmap([[ 0., 0., 0., 0.],
[ 0., 0., 0., 0.],
[ 0., 0., 0., 0.]], dtype=float32)

Write data to memmap array:

>>> fp[:] = data[:]
>>> fp
memmap([[ 0., 1., 2., 3.],
[ 4., 5., 6., 7.],
[ 8., 9., 10., 11.]], dtype=float32)

>>> fp.filename == path.abspath(filename)

True

Deletion flushes memory changes to disk before removing the object:

>>> del fp

Load the memmap and verify data was stored:

>>> newfp = np.memmap(filename, dtype=’float32’, mode=’r’, shape=(3,4))
>>> newfp
memmap([[ 0., 1., 2., 3.],
[ 4., 5., 6., 7.],
[ 8., 9., 10., 11.]], dtype=float32)

Read-only memmap:
>>> fpr = np.memmap(filename, dtype=’float32’, mode=’r’, shape=(3,4))
>>> fpr.flags.writeable
False

Copy-on-write memmap:
>>> fpc = np.memmap(filename, dtype=’float32’, mode=’c’, shape=(3,4))
>>> fpc.flags.writeable
True

It’s possible to assign to copy-on-write array, but values are only written into the memory copy of the array, and
not written to disk:
>>> fpc
memmap([[ 0., 1., 2., 3.],
[ 4., 5., 6., 7.],
[ 8., 9., 10., 11.]], dtype=float32)
>>> fpc[0,:] = 0
>>> fpc
memmap([[ 0., 0., 0., 0.],
[ 4., 5., 6., 7.],
[ 8., 9., 10., 11.]], dtype=float32)

File on disk is unchanged:

>>> fpr
memmap([[ 0., 1., 2., 3.],

166 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

[ 4., 5., 6., 7.],

[ 8., 9., 10., 11.]], dtype=float32)

Offset into a memmap:

>>> fpo = np.memmap(filename, dtype=’float32’, mode=’r’, offset=16)
>>> fpo
memmap([ 4., 5., 6., 7., 8., 9., 10., 11.], dtype=float32)

Attributes
filename str Path to the mapped file.
offset int Offset position in the file.
mode str File mode.

Methods

memmap.flush()
Write any changes in the array to the file on disk.
For further information, see memmap.
Parameters
None :
See Also:
memmap
Example:
>>> a = memmap(’newfile.dat’, dtype=float, mode=’w+’, shape=1000)
>>> a[10] = 10.0
>>> a[30] = 30.0
>>> del a
>>> b = fromfile(’newfile.dat’, dtype=float)
>>> print b[10], b[30]
10.0 30.0
>>> a = memmap(’newfile.dat’, dtype=float)
>>> print a[10], a[30]
10.0 30.0

1.5.4 Character arrays (numpy.char)

See Also:
Creating character arrays (numpy.char)
Note: The chararray class exists for backwards compatibility with Numarray, it is not recommended for new devel-
opment. Starting from numpy 1.4, if one needs arrays of strings, it is recommended to use arrays of dtype object_,
string_ or unicode_, and use the free functions in the numpy.char module for fast vectorized string operations.
These are enhanced arrays of either string_ type or unicode_ type. These arrays inherit from the ndarray,
but specially-define the operations +, *, and % on a (broadcasting) element-by-element basis. These operations are
not available on the standard ndarray of character type. In addition, the chararray has all of the standard
string (and unicode) methods, executing them on an element-by-element basis. Perhaps the easiest way to create
a chararray is to use self.view(chararray) where self is an ndarray of str or unicode data-type. However, a
chararray can also be created using the numpy.chararray constructor, or via the numpy.char.array function:

1.5. Standard array subclasses 167

NumPy Reference, Release 2.0.0.dev-ac3cba3

chararray Provides a convenient view on arrays of string and unicode

values.
core.defchararray.array(obj[, itemsize, Create a chararray.
...])
class numpy.chararray
Provides a convenient view on arrays of string and unicode values.
Note: The chararray class exists for backwards compatibility with Numarray, it is not recommended for new
development. Starting from numpy 1.4, if one needs arrays of strings, it is recommended to use arrays of dtype
object_, string_ or unicode_, and use the free functions in the numpy.char module for fast vectorized string
operations.
Versus a regular Numpy array of type str or unicode, this class adds the following functionality:
1.values automatically have whitespace removed from the end when indexed
2.comparison operators automatically remove whitespace from the end when comparing values
3.vectorized string operations are provided as methods (e.g. endswith) and infix operators (e.g. "+", "*",
"%")
chararrays should be created using numpy.char.array or numpy.char.asarray, rather than this con-
structor directly.
This constructor creates the array, using buffer (with offset and strides) if it is not None. If buffer is None, then
constructs a new array with strides in “C order”, unless both len(shape) >= 2 and order=’Fortran’,
in which case strides is in “Fortran order”.
Parameters
shape : tuple
Shape of the array.
itemsize : int, optional
Length of each array element, in number of characters. Default is 1.
unicode : bool, optional
Are the array elements of type unicode (True) or string (False). Default is False.
buffer : int, optional
Memory address of the start of the array data. Default is None, in which case a new
array is created.
offset : int, optional
Fixed stride displacement from the beginning of an axis? Default is 0. Needs to be >=0.
strides : array_like of ints, optional
Strides for the array (see ndarray.strides for full description). Default is None.
order : {‘C’, ‘F’}, optional
The order in which the array data is stored in memory: ‘C’ -> “row major” order (the
default), ‘F’ -> “column major” (Fortran) order.

Examples
>>> charar = np.chararray((3, 3))
>>> charar[:] = ’a’
>>> charar
chararray([[’a’, ’a’, ’a’],

168 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

[’a’, ’a’, ’a’],

[’a’, ’a’, ’a’]],
dtype=’|S1’)

>>> charar = np.chararray(charar.shape, itemsize=5)

>>> charar[:] = ’abc’
>>> charar
chararray([[’abc’, ’abc’, ’abc’],
[’abc’, ’abc’, ’abc’],
[’abc’, ’abc’, ’abc’]],
dtype=’|S5’)

Methods

astype
argsort(a[, axis, kind, order]) Returns the indices that would sort an array.
copy(a) Return an array copy of the given object.
count
decode
dump
dumps
encode
endswith
expandtabs
fill
find
flatten
getfield
index
isalnum
isalpha
isdecimal
isdigit
islower
isnumeric
isspace
istitle
isupper
item
join
ljust
lower
lstrip
nonzero(a) Return the indices of the elements that are non-zero.
put(a, ind, v[, mode]) Replaces specified elements of an array with given values.
ravel(a[, order]) Return a flattened array.
repeat(a, repeats[, axis]) Repeat elements of an array.
replace
reshape(a, newshape[, order]) Gives a new shape to an array without changing its data.
resize(a, new_shape) Return a new array with the specified shape.
rfind
rindex
rjust
Continued on next page

1.5. Standard array subclasses 169

NumPy Reference, Release 2.0.0.dev-ac3cba3

Table 1.4 – continued from previous page

rsplit
rstrip
searchsorted(a, v[, side]) Find indices where elements should be inserted to maintain order.
setfield
setflags
sort(a[, axis, kind, order]) Return a sorted copy of an array.
split(ary, indices_or_sections[, axis]) Split an array into multiple sub-arrays of equal size.
splitlines
squeeze(a) Remove single-dimensional entries from the shape of an array.
startswith
strip
swapaxes(a, axis1, axis2) Interchange two axes of an array.
swapcase
take(a, indices[, axis, out, mode]) Take elements from an array along an axis.
title
tofile
tolist
tostring
translate
transpose(a[, axes]) Permute the dimensions of an array.
upper
view
zfill

numpy.argsort(a, axis=-1, kind=’quicksort’, order=None)

170 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

sort
Describes sorting algorithms used.
lexsort
Indirect stable sort with multiple keys.
ndarray.sort
Inplace sort.

Notes
See sort for notes on the different sorting algorithms.
As of NumPy 1.4.0 argsort works with real/complex arrays containing nan values. The enhanced sort
order is documented in sort.

Examples
One dimensional array:
>>> x = np.array([3, 1, 2])
>>> np.argsort(x)
array([1, 2, 0])

Two-dimensional array:
>>> x = np.array([[0, 3], [2, 2]])
>>> x
array([[0, 3],
[2, 2]])

>>> np.argsort(x, axis=0)

array([[0, 1],
[1, 0]])

>>> np.argsort(x, axis=1)

array([[0, 1],
[0, 1]])

Sorting with keys:

>>> x = np.array([(1, 0), (0, 1)], dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])
>>> x
array([(1, 0), (0, 1)],
dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])

>>> np.argsort(x, order=(’x’,’y’))

array([1, 0])

>>> np.argsort(x, order=(’y’,’x’))

array([0, 1])

numpy.copy(a)
Return an array copy of the given object.
Parameters
a : array_like
Input data.
Returns
arr : ndarray

1.5. Standard array subclasses 171

NumPy Reference, Release 2.0.0.dev-ac3cba3

Array interpretation of a.

Notes
This is equivalent to
>>> np.array(a, copy=True)

Examples
Create an array x, with a reference y and a copy z:
>>> x = np.array([1, 2, 3])
>>> y = x
>>> z = np.copy(x)

Note that, when we modify x, y changes, but not z:

>>> x[0] = 10
>>> x[0] == y[0]
True
>>> x[0] == z[0]
False

To group the indices by element, rather than dimension, use:

transpose(nonzero(a))

The result of this is always a 2-D array, with a row for each non-zero element.
Parameters
a : array_like
Input array.
Returns
tuple_of_arrays : tuple
Indices of elements that are non-zero.
See Also:

Examples

172 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x = np.eye(3)
>>> x
array([[ 1., 0., 0.],
[ 0., 1., 0.],
[ 0., 0., 1.]])
>>> np.nonzero(x)
(array([0, 1, 2]), array([0, 1, 2]))

>>> x[np.nonzero(x)]
array([ 1., 1., 1.])
>>> np.transpose(np.nonzero(x))
array([[0, 0],
[1, 1],
[2, 2]])

The nonzero method of the boolean array can also be called.

>>> (a > 3).nonzero()
(array([1, 1, 1, 2, 2, 2]), array([0, 1, 2, 0, 1, 2]))

numpy.put(a, ind, v, mode=’raise’)

Replaces specified elements of an array with given values.
The indexing works on the flattened target array. put is roughly equivalent to:
a.flat[ind] = v

1.5. Standard array subclasses 173

NumPy Reference, Release 2.0.0.dev-ac3cba3

‘clip’ mode means that all indices that are too large are replaced by the index that ad-
dresses the last element along that axis. Note that this disables indexing with negative
numbers.

See Also:
putmask, place

Examples
>>> a = np.arange(5)
>>> np.put(a, [0, 2], [-44, -55])
>>> a
array([-44, 1, -55, 3, 4])

>>> a = np.arange(5)
>>> np.put(a, 22, -5, mode=’clip’)
>>> a
array([ 0, 1, 2, 3, -5])

ndarray.flat
1-D iterator over an array.
ndarray.flatten
1-D array copy of the elements of an array in row-major order.

174 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
It is equivalent to reshape(-1, order=order).
>>> x = np.array([[1, 2, 3], [4, 5, 6]])
>>> print np.ravel(x)
[1 2 3 4 5 6]

>>> print x.reshape(-1)

[1 2 3 4 5 6]

>>> print np.ravel(x, order=’F’)

[1 4 2 5 3 6]

When order is ‘A’, it will preserve the array’s ‘C’ or ‘F’ ordering:
>>> print np.ravel(x.T)
[1 4 2 5 3 6]
>>> print np.ravel(x.T, order=’A’)
[1 2 3 4 5 6]

numpy.repeat(a, repeats, axis=None)

1.5. Standard array subclasses 175

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:

tile
Tile an array.

numpy.reshape(a, newshape, order=’C’)

ndarray.reshape
Equivalent method.

176 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> a = np.array([[1,2,3], [4,5,6]])
>>> np.reshape(a, 6)
array([1, 2, 3, 4, 5, 6])
>>> np.reshape(a, 6, order=’F’)
array([1, 4, 2, 5, 3, 6])

>>> np.reshape(a, (3,-1)) # the unspecified value is inferred to be 2

array([[1, 2],
[3, 4],
[5, 6]])

ndarray.resize
resize an array in-place.

Examples
>>> a=np.array([[0,1],[2,3]])
>>> np.resize(a,(1,4))
array([[0, 1, 2, 3]])
>>> np.resize(a,(2,4))
array([[0, 1, 2, 3],
[0, 1, 2, 3]])

1.5. Standard array subclasses 177

NumPy Reference, Release 2.0.0.dev-ac3cba3

Values to insert into a.

side : {‘left’, ‘right’}, optional
If ‘left’, the index of the first suitable location found is given. If ‘right’, return the last
such index. If there is no suitable index, return either 0 or N (where N is the length of
a).
Returns
indices : array of ints
Array of insertion points with the same shape as v.
See Also:

sort
Return a sorted copy of an array.
histogram
Produce histogram from 1-D data.

Notes
Binary search is used to find the required insertion points.
As of Numpy 1.4.0 searchsorted works with real/complex arrays containing nan values. The enhanced
sort order is documented in sort.

Examples
>>> np.searchsorted([1,2,3,4,5], 3)
2
>>> np.searchsorted([1,2,3,4,5], 3, side=’right’)
3
>>> np.searchsorted([1,2,3,4,5], [-10, 10, 2, 3])
array([0, 5, 1, 2])

numpy.sort(a, axis=-1, kind=’quicksort’, order=None)

178 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.sort
Method to sort an array in-place.
argsort
Indirect sort.
lexsort
Indirect stable sort on multiple keys.
searchsorted
Find elements in a sorted array.

1.5. Standard array subclasses 179

NumPy Reference, Release 2.0.0.dev-ac3cba3

Sort by age, then height if ages are equal:

numpy.split(ary, indices_or_sections, axis=0)

Split an array into multiple sub-arrays of equal size.
Parameters
ary : ndarray
Array to be divided into sub-arrays.
indices_or_sections : int or 1-D array
If indices_or_sections is an integer, N, the array will be divided into N equal arrays
along axis. If such a split is not possible, an error is raised.
If indices_or_sections is a 1-D array of sorted integers, the entries indicate where along
axis the array is split. For example, [2, 3] would, for axis=0, result in
• ary[:2]
• ary[2:3]
• ary[3:]
If an index exceeds the dimension of the array along axis, an empty sub-array is returned
correspondingly.
axis : int, optional
The axis along which to split, default is 0.
Returns
sub-arrays : list of ndarrays
A list of sub-arrays.
Raises
ValueError :
If indices_or_sections is given as an integer, but a split does not result in equal division.
See Also:

array_split
Split an array into multiple sub-arrays of equal or near-equal size. Does not raise an exception if an
equal division cannot be made.
hsplit
Split array into multiple sub-arrays horizontally (column-wise).
vsplit
Split array into multiple sub-arrays vertically (row wise).
dsplit
Split array into multiple sub-arrays along the 3rd axis (depth).
concatenate
Join arrays together.

180 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

hstack
Stack arrays in sequence horizontally (column wise).
vstack
Stack arrays in sequence vertically (row wise).
dstack
Stack arrays in sequence depth wise (along third dimension).

Examples
>>> x = np.arange(9.0)
>>> np.split(x, 3)
[array([ 0., 1., 2.]), array([ 3., 4., 5.]), array([ 6., 7., 8.])]

>>> x = np.arange(8.0)
>>> np.split(x, [3, 5, 6, 10])
[array([ 0., 1., 2.]),
array([ 3., 4.]),
array([ 5.]),
array([ 6., 7.]),
array([], dtype=float64)]

Examples
>>> x = np.array([[[0], [1], [2]]])
>>> x.shape
(1, 3, 1)
>>> np.squeeze(x).shape
(3,)

numpy.swapaxes(a, axis1, axis2)

Interchange two axes of an array.
Parameters
a : array_like
Input array.
axis1 : int
First axis.
axis2 : int
Second axis.
Returns
a_swapped : ndarray

1.5. Standard array subclasses 181

NumPy Reference, Release 2.0.0.dev-ac3cba3

If a is an ndarray, then a view of a is returned; otherwise a new array is created.

Examples
>>> x = np.array([[1,2,3]])
>>> np.swapaxes(x,0,1)
array([[1],
[2],
[3]])

>>> x = np.array([[[0,1],[2,3]],[[4,5],[6,7]]])
>>> x
array([[[0, 1],
[2, 3]],
[[4, 5],
[6, 7]]])

>>> np.swapaxes(x,0,2)
array([[[0, 4],
[2, 6]],
[[1, 5],
[3, 7]]])

numpy.take(a, indices, axis=None, out=None, mode=’raise’)

182 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:

ndarray.take
equivalent method

Examples
>>> a = [4, 3, 5, 7, 6, 8]
>>> indices = [0, 1, 4]
>>> np.take(a, indices)
array([4, 3, 6])

In this example if a is an ndarray, “fancy” indexing can be used.

>>> a = np.array(a)
>>> a[indices]
array([4, 3, 6])

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.transpose(x)
array([[0, 2],
[1, 3]])

>>> x = np.ones((1, 2, 3))

>>> np.transpose(x, (1, 0, 2)).shape
(2, 1, 3)

numpy.core.defchararray.array(obj, itemsize=None, copy=True, unicode=None, order=None)

Create a chararray.
Note: This class is provided for numarray backward-compatibility. New code (not concerned with numarray
compatibility) should use arrays of type string_ or unicode_ and use the free functions in numpy.char for fast
vectorized string operations instead.

1.5. Standard array subclasses 183

NumPy Reference, Release 2.0.0.dev-ac3cba3

Versus a regular Numpy array of type str or unicode, this class adds the following functionality:
1.values automatically have whitespace removed from the end when indexed
2.comparison operators automatically remove whitespace from the end when comparing values
3.vectorized string operations are provided as methods (e.g. str.endswith) and infix operators (e.g. +, *,
%)

Parameters
obj : array of str or unicode-like
itemsize : int, optional
itemsize is the number of characters per scalar in the resulting array. If itemsize is
None, and obj is an object array or a Python list, the itemsize will be automatically
determined. If itemsize is provided and obj is of type str or unicode, then the obj string
will be chunked into itemsize pieces.
copy : bool, optional
If true (default), then the object is copied. Otherwise, a copy will only be made if
__array__ returns a copy, if obj is a nested sequence, or if a copy is needed to satisfy
any of the other requirements (itemsize, unicode, order, etc.).
unicode : bool, optional
When true, the resulting chararray can contain Unicode characters, when false only
8-bit characters. If unicode is None and obj is one of the following:
• a chararray,
• an ndarray of type str or unicode
• a Python str or unicode object,
then the unicode setting of the output array will be automatically determined.
order : {‘C’, ‘F’, ‘A’}, optional
Specify the order of the array. If order is ‘C’ (default), then the array will be in C-
contiguous order (last-index varies the fastest). If order is ‘F’, then the returned array
will be in Fortran-contiguous order (first-index varies the fastest). If order is ‘A’, then
the returned array may be in any order (either C-, Fortran-contiguous, or even discon-
tiguous).

Another difference with the standard ndarray of str data-type is that the chararray inherits the feature introduced by
Numarray that white-space at the end of any element in the array will be ignored on item retrieval and comparison
operations.

1.5.5 Record arrays (numpy.rec)

See Also:
Creating record arrays (numpy.rec), Data type routines, Data type objects (dtype).
Numpy provides the recarray class which allows accessing the fields of a record/structured array as attributes, and
a corresponding scalar data type object record.
recarray Construct an ndarray that allows field access using attributes.
record A data-type scalar that allows field access as attribute lookup.

184 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

class numpy.recarray
Construct an ndarray that allows field access using attributes.
Arrays may have a data-types containing fields, analogous to columns in a spread sheet. An example is [(x,
int), (y, float)], where each entry in the array is a pair of (int, float). Normally, these attributes
are accessed using dictionary lookups such as arr[’x’] and arr[’y’]. Record arrays allow the fields to be
accessed as members of the array, using arr.x and arr.y.
Parameters
shape : tuple
Shape of output array.
dtype : data-type, optional
The desired data-type. By default, the data-type is determined from formats, names,
titles, aligned and byteorder.
formats : list of data-types, optional
A list containing the data-types for the different columns, e.g. [’i4’, ’f8’,
’i4’]. formats does not support the new convention of using types directly, i.e.
(int, float, int). Note that formats must be a list, not a tuple. Given that
formats is somewhat limited, we recommend specifying dtype instead.
names : tuple of str, optional
The name of each column, e.g. (’x’, ’y’, ’z’).
buf : buffer, optional
By default, a new array is created of the given shape and data-type. If buf is specified
and is an object exposing the buffer interface, the array will use the memory from the
existing buffer. In this case, the offset and strides keywords are available.
Returns
rec : recarray
Empty array of the given shape and type.
Other Parameters
titles : tuple of str, optional
Aliases for column names. For example, if names were (’x’, ’y’, ’z’) and
titles is (’x_coordinate’, ’y_coordinate’, ’z_coordinate’), then
arr[’x’] is equivalent to both arr.x and arr.x_coordinate.
byteorder : {‘<’, ‘>’, ‘=’}, optional
Byte-order for all fields.
aligned : bool, optional
Align the fields in memory as the C-compiler would.
strides : tuple of ints, optional
Buffer (buf ) is interpreted according to these strides (strides define how many bytes
each array element, row, column, etc. occupy in memory).
offset : int, optional
Start reading buffer (buf ) from this offset onwards.
order : {‘C’, ‘F’}, optional

1.5. Standard array subclasses 185

NumPy Reference, Release 2.0.0.dev-ac3cba3

Row-major or column-major order.

See Also:

rec.fromrecords
Construct a record array from data.
record
fundamental data-type for recarray.
format_parser
determine a data-type from formats, names, titles.

Notes
This constructor can be compared to empty: it creates a new record array but does not fill it with data. To create
a record array from data, use one of the following methods:
1.Create a standard ndarray and convert it to a record array, using arr.view(np.recarray)
2.Use the buf keyword.
3.Use np.rec.fromrecords.

Examples
Create an array with two fields, x and y:
>>> x = np.array([(1.0, 2), (3.0, 4)], dtype=[(’x’, float), (’y’, int)])
>>> x
array([(1.0, 2), (3.0, 4)],
dtype=[(’x’, ’<f8’), (’y’, ’<i4’)])

>>> x[’x’]
array([ 1., 3.])

View the array as a record array:

>>> x = x.view(np.recarray)

>>> x.x
array([ 1., 3.])

>>> x.y
array([2, 4])

Create a new, empty record array:

>>> np.recarray((2,),
... dtype=[(’x’, int), (’y’, float), (’z’, int)])
rec.array([(-1073741821, 1.2249118382103472e-301, 24547520),
(3471280, 1.2134086255804012e-316, 0)],
dtype=[(’x’, ’<i4’), (’y’, ’<f8’), (’z’, ’<i4’)])

Methods

186 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Table 1.5 – continued from previous page

argsort(a[, axis, kind, order]) Returns the indices that would sort an array.
astype
byteswap
choose(a, choices[, out, mode]) Construct an array from an index array and a set of arrays to choose from.
clip(a, a_min, a_max[, out]) Clip (limit) the values in an array.
compress(condition, a[, axis, out]) Return selected slices of an array along given axis.
conj() Return the complex conjugate, element-wise.
conjugate() Return the complex conjugate, element-wise.
copy(a) Return an array copy of the given object.
cumprod(a[, axis, dtype, out]) Return the cumulative product of elements along a given axis.
cumsum(a[, axis, dtype, out]) Return the cumulative sum of the elements along a given axis.
diagonal(a[, offset, axis1, axis2]) Return specified diagonals.
dot(a, b[, out]) Dot product of two arrays.
dump
dumps
field
fill
flatten
getfield
item
itemset
max(a[, axis, out]) Return the maximum of an array or maximum along an axis.
mean(a[, axis, dtype, out]) Compute the arithmetic mean along the specified axis.
min(a[, axis, out]) Return the minimum of an array or minimum along an axis.
newbyteorder
nonzero(a) Return the indices of the elements that are non-zero.
prod(a[, axis, dtype, out]) Return the product of array elements over a given axis.
ptp(a[, axis, out]) Range of values (maximum - minimum) along an axis.
put(a, ind, v[, mode]) Replaces specified elements of an array with given values.
ravel(a[, order]) Return a flattened array.
repeat(a, repeats[, axis]) Repeat elements of an array.
reshape(a, newshape[, order]) Gives a new shape to an array without changing its data.
resize(a, new_shape) Return a new array with the specified shape.
round(a[, decimals, out]) Round an array to the given number of decimals.
searchsorted(a, v[, side]) Find indices where elements should be inserted to maintain order.
setasflat
setfield
setflags
sort(a[, axis, kind, order]) Return a sorted copy of an array.
squeeze(a) Remove single-dimensional entries from the shape of an array.
std(a[, axis, dtype, out, ddof]) Compute the standard deviation along the specified axis.
sum(a[, axis, dtype, out]) Sum of array elements over a given axis.
swapaxes(a, axis1, axis2) Interchange two axes of an array.
take(a, indices[, axis, out, mode]) Take elements from an array along an axis.
tofile
tolist
tostring
trace(a[, offset, axis1, axis2, dtype, out]) Return the sum along diagonals of the array.
transpose(a[, axes]) Permute the dimensions of an array.
var(a[, axis, dtype, out, ddof]) Compute the variance along the specified axis.
view

1.5. Standard array subclasses 187

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.all(a, axis=None, out=None)

ndarray.all
equivalent method
any
Test whether any element along a given axis evaluates to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.all([[True,False],[True,True]])
False

>>> np.all([[True,False],[True,True]], axis=0)

array([ True, False], dtype=bool)

>>> np.all([-1, 4, 5])

True

>>> np.all([1.0, np.nan])

True

>>> o=np.array([False])
>>> z=np.all([-1, 4, 5], out=o)
>>> id(z), id(o), z
(28293632, 28293632, array([ True], dtype=bool))

numpy.any(a, axis=None, out=None)

Test whether any array element along a given axis evaluates to True.

188 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Returns single boolean unless axis is not None

Parameters
a : array_like
Input array or object that can be converted to an array.
axis : int, optional
Axis along which a logical OR is performed. The default (axis = None) is to perform
a logical OR over a flattened input array. axis may be negative, in which case it counts
from the last to the first axis.
out : ndarray, optional
Alternate output array in which to place the result. It must have the same shape as the
expected output and its type is preserved (e.g., if it is of type float, then it will remain
so, returning 1.0 for True and 0.0 for False, regardless of the type of a). See doc.ufuncs
(Section “Output arguments”) for details.
Returns
any : bool or ndarray
A new boolean or ndarray is returned unless out is specified, in which case a reference
to out is returned.
See Also:

ndarray.any
equivalent method
all
Test whether all elements along a given axis evaluate to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.any([[True, False], [True, True]])
True

>>> np.any([[True, False], [False, False]], axis=0)

array([ True, False], dtype=bool)

>>> np.any([-1, 0, 5])

True

>>> np.any(np.nan)
True

1.5. Standard array subclasses 189

NumPy Reference, Release 2.0.0.dev-ac3cba3

Notes
In case of multiple occurrences of the maximum values, the indices corresponding to the first occurrence
are returned.

Examples
>>> a = np.arange(6).reshape(2,3)
>>> a
array([[0, 1, 2],
[3, 4, 5]])
>>> np.argmax(a)
5
>>> np.argmax(a, axis=0)
array([1, 1, 1])
>>> np.argmax(a, axis=1)
array([2, 2])

>>> b = np.arange(6)
>>> b[1] = 5
>>> b
array([0, 5, 2, 3, 4, 5])
>>> np.argmax(b) # Only the first occurrence is returned.
1

numpy.argmin(a, axis=None)
Return the indices of the minimum values along an axis.
See Also:

argmax
Similar function. Please refer to numpy.argmax for detailed documentation.

190 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.argsort(a, axis=-1, kind=’quicksort’, order=None)

sort
Describes sorting algorithms used.
lexsort
Indirect stable sort with multiple keys.
ndarray.sort
Inplace sort.

Notes
See sort for notes on the different sorting algorithms.
As of NumPy 1.4.0 argsort works with real/complex arrays containing nan values. The enhanced sort
order is documented in sort.

Examples
One dimensional array:
>>> x = np.array([3, 1, 2])
>>> np.argsort(x)
array([1, 2, 0])

Two-dimensional array:
>>> x = np.array([[0, 3], [2, 2]])
>>> x
array([[0, 3],
[2, 2]])

1.5. Standard array subclasses 191

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.argsort(x, axis=0)

array([[0, 1],
[1, 0]])

>>> np.argsort(x, axis=1)

array([[0, 1],
[0, 1]])

Sorting with keys:

>>> x = np.array([(1, 0), (0, 1)], dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])
>>> x
array([(1, 0), (0, 1)],
dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])

>>> np.argsort(x, order=(’x’,’y’))

array([1, 0])

>>> np.argsort(x, order=(’y’,’x’))

array([0, 1])

numpy.choose(a, choices, out=None, mode=’raise’)

192 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

If provided, the result will be inserted into this array. It should be of the appropriate
shape and dtype.
mode : {‘raise’ (default), ‘wrap’, ‘clip’}, optional
Specifies how indices outside [0, n-1] will be treated:
• ‘raise’ : an exception is raised
• ‘wrap’ : value becomes value mod n
• ‘clip’ : values < 0 are mapped to 0, values > n-1 are mapped to n-1
Returns
merged_array : array
The merged result.
Raises
ValueError: shape mismatch :
If a and each choice array are not all broadcastable to the same shape.

See Also:

ndarray.choose
equivalent method

A couple examples illustrating how choose broadcasts:

>>> a = [[1, 0, 1], [0, 1, 0], [1, 0, 1]]
>>> choices = [-10, 10]
>>> np.choose(a, choices)
array([[ 10, -10, 10],
[-10, 10, -10],
[ 10, -10, 10]])

1.5. Standard array subclasses 193

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> # With thanks to Anne Archibald

numpy.clip(a, a_min, a_max, out=None)

numpy.doc.ufuncs
Section “Output arguments”

194 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.compress(condition, a, axis=None, out=None)

Working on the flattened array does not return slices along an axis but selects elements.
>>> np.compress([False, True], a)
array([2])

numpy.conj(x[, out ])
Return the complex conjugate, element-wise.
The complex conjugate of a complex number is obtained by changing the sign of its imaginary part.

1.5. Standard array subclasses 195

NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
x : array_like
Input value.
Returns
y : ndarray
The complex conjugate of x, with same dtype as y.

Examples
>>> np.conjugate(1+2j)
(1-2j)

>>> x = np.eye(2) + 1j * np.eye(2)

>>> np.conjugate(x)
array([[ 1.-1.j, 0.-0.j],
[ 0.-0.j, 1.-1.j]])

numpy.copy(a)
Return an array copy of the given object.
Parameters
a : array_like
Input data.
Returns
arr : ndarray
Array interpretation of a.

Notes
This is equivalent to
>>> np.array(a, copy=True)

Examples
Create an array x, with a reference y and a copy z:
>>> x = np.array([1, 2, 3])
>>> y = x
>>> z = np.copy(x)

Note that, when we modify x, y changes, but not z:

>>> x[0] = 10
>>> x[0] == y[0]
True
>>> x[0] == z[0]
False

numpy.cumprod(a, axis=None, dtype=None, out=None)

Return the cumulative product of elements along a given axis.
Parameters
a : array_like
Input array.

196 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

axis : int, optional

numpy.doc.ufuncs
Section “Output arguments”

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

The cumulative product for each column (i.e., over the rows) of a:
>>> np.cumprod(a, axis=0)
array([[ 1, 2, 3],
[ 4, 10, 18]])

The cumulative product for each row (i.e. over the columns) of a:
>>> np.cumprod(a,axis=1)
array([[ 1, 2, 6],
[ 4, 20, 120]])

numpy.cumsum(a, axis=None, dtype=None, out=None)

Return the cumulative sum of the elements along a given axis.
Parameters
a : array_like
Input array.
axis : int, optional

1.5. Standard array subclasses 197

NumPy Reference, Release 2.0.0.dev-ac3cba3

sum
Sum array elements.
trapz
Integration of array values using the composite trapezoidal rule.

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

>>> np.cumsum(a,axis=0) # sum over rows for each of the 3 columns

array([[1, 2, 3],
[5, 7, 9]])
>>> np.cumsum(a,axis=1) # sum over columns for each of the 2 rows
array([[ 1, 3, 6],
[ 4, 9, 15]])

numpy.diagonal(a, offset=0, axis1=0, axis2=1)

198 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

diag
MATLAB work-a-like for 1-D and 2-D arrays.
diagflat
Create diagonal arrays.
trace
Sum along diagonals.

Examples
>>> a = np.arange(4).reshape(2,2)
>>> a
array([[0, 1],
[2, 3]])
>>> a.diagonal()
array([0, 3])
>>> a.diagonal(1)
array([1])

1.5. Standard array subclasses 199

NumPy Reference, Release 2.0.0.dev-ac3cba3

... 1) # the "middle" (row) axis first.

array([[0, 6],
[1, 7]])

See Also:

vdot
Complex-conjugating dot product.
tensordot
Sum products over arbitrary axes.
einsum
Einstein summation convention.

200 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> np.dot(3, 4)
12

Neither argument is complex-conjugated:

>>> np.dot([2j, 3j], [2j, 3j])
(-13+0j)

For 2-D arrays it’s the matrix product:

>>> a = [[1, 0], [0, 1]]
>>> b = [[4, 1], [2, 2]]
>>> np.dot(a, b)
array([[4, 1],
[2, 2]])

>>> a = np.arange(3*4*5*6).reshape((3,4,5,6))
>>> b = np.arange(3*4*5*6)[::-1].reshape((5,4,6,3))
>>> np.dot(a, b)[2,3,2,1,2,2]
499128
>>> sum(a[2,3,2,:] * b[1,2,:,2])
499128

numpy.mean(a, axis=None, dtype=None, out=None)

average
Weighted average

1.5. Standard array subclasses 201

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.mean(a)
2.5
>>> np.mean(a, axis=0)
array([ 2., 3.])
>>> np.mean(a, axis=1)
array([ 1.5, 3.5])

In single precision, mean can be inaccurate:

>>> a = np.zeros((2, 512*512), dtype=np.float32)
>>> a[0, :] = 1.0
>>> a[1, :] = 0.1
>>> np.mean(a)
0.546875

Computing the mean in float64 is more accurate:

>>> np.mean(a, dtype=np.float64)
0.55000000074505806

To group the indices by element, rather than dimension, use:

transpose(nonzero(a))

The result of this is always a 2-D array, with a row for each non-zero element.
Parameters
a : array_like
Input array.
Returns
tuple_of_arrays : tuple
Indices of elements that are non-zero.
See Also:

flatnonzero
Return indices that are non-zero in the flattened version of the input array.
ndarray.nonzero
Equivalent ndarray method.

202 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

count_nonzero
Counts the number of non-zero elements in the input array.

Examples
>>> x = np.eye(3)
>>> x
array([[ 1., 0., 0.],
[ 0., 1., 0.],
[ 0., 0., 1.]])
>>> np.nonzero(x)
(array([0, 1, 2]), array([0, 1, 2]))

>>> x[np.nonzero(x)]
array([ 1., 1., 1.])
>>> np.transpose(np.nonzero(x))
array([[0, 0],
[1, 1],
[2, 2]])

The nonzero method of the boolean array can also be called.

>>> (a > 3).nonzero()
(array([1, 1, 1, 2, 2, 2]), array([0, 1, 2, 0, 1, 2]))

numpy.prod(a, axis=None, dtype=None, out=None)

1.5. Standard array subclasses 203

NumPy Reference, Release 2.0.0.dev-ac3cba3

Returns
product_along_axis : ndarray, see dtype parameter above.
An array shaped as a but with the specified axis removed. Returns a reference to out if
specified.
See Also:

ndarray.prod
equivalent method
numpy.doc.ufuncs
Section “Output arguments”

Examples
By default, calculate the product of all elements:
>>> np.prod([1.,2.])
2.0

Even when the input array is two-dimensional:

>>> np.prod([[1.,2.],[3.,4.]])
24.0

But we can also specify the axis over which to multiply:

>>> np.prod([[1.,2.],[3.,4.]], axis=1)
array([ 2., 12.])

If the type of x is unsigned, then the output type is the unsigned platform integer:
>>> x = np.array([1, 2, 3], dtype=np.uint8)
>>> np.prod(x).dtype == np.uint
True

If x is of a signed integer type, then the output type is the default platform integer:
>>> x = np.array([1, 2, 3], dtype=np.int8)
>>> np.prod(x).dtype == np.int
True

numpy.ptp(a, axis=None, out=None)

Range of values (maximum - minimum) along an axis.
The name of the function comes from the acronym for ‘peak to peak’.
Parameters
a : array_like
Input values.
axis : int, optional

204 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Axis along which to find the peaks. By default, flatten the array.
out : array_like
Alternative output array in which to place the result. It must have the same shape and
buffer length as the expected output, but the type of the output values will be cast if
necessary.
Returns
ptp : ndarray
A new array holding the result, unless out was specified, in which case a reference to
out is returned.

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.ptp(x, axis=0)

array([2, 2])

>>> np.ptp(x, axis=1)

array([1, 1])

numpy.put(a, ind, v, mode=’raise’)

Replaces specified elements of an array with given values.
The indexing works on the flattened target array. put is roughly equivalent to:
a.flat[ind] = v

1.5. Standard array subclasses 205

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:
putmask, place

Examples
>>> a = np.arange(5)
>>> np.put(a, [0, 2], [-44, -55])
>>> a
array([-44, 1, -55, 3, 4])

>>> a = np.arange(5)
>>> np.put(a, 22, -5, mode=’clip’)
>>> a
array([ 0, 1, 2, 3, -5])

ndarray.flat
1-D iterator over an array.
ndarray.flatten
1-D array copy of the elements of an array in row-major order.

Examples
It is equivalent to reshape(-1, order=order).
>>> x = np.array([[1, 2, 3], [4, 5, 6]])
>>> print np.ravel(x)
[1 2 3 4 5 6]

206 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> print x.reshape(-1)

[1 2 3 4 5 6]

>>> print np.ravel(x, order=’F’)

[1 4 2 5 3 6]

When order is ‘A’, it will preserve the array’s ‘C’ or ‘F’ ordering:
>>> print np.ravel(x.T)
[1 4 2 5 3 6]
>>> print np.ravel(x.T, order=’A’)
[1 2 3 4 5 6]

numpy.repeat(a, repeats, axis=None)

tile
Tile an array.

1.5. Standard array subclasses 207

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.reshape(a, newshape, order=’C’)

ndarray.reshape
Equivalent method.

Examples
>>> a = np.array([[1,2,3], [4,5,6]])
>>> np.reshape(a, 6)
array([1, 2, 3, 4, 5, 6])

208 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.reshape(a, 6, order=’F’)

array([1, 4, 2, 5, 3, 6])

>>> np.reshape(a, (3,-1)) # the unspecified value is inferred to be 2

array([[1, 2],
[3, 4],
[5, 6]])

ndarray.resize
resize an array in-place.

Examples
>>> a=np.array([[0,1],[2,3]])
>>> np.resize(a,(1,4))
array([[0, 1, 2, 3]])
>>> np.resize(a,(2,4))
array([[0, 1, 2, 3],
[0, 1, 2, 3]])

1.5. Standard array subclasses 209

NumPy Reference, Release 2.0.0.dev-ac3cba3

If ‘left’, the index of the first suitable location found is given. If ‘right’, return the last
such index. If there is no suitable index, return either 0 or N (where N is the length of
a).
Returns
indices : array of ints
Array of insertion points with the same shape as v.
See Also:

sort
Return a sorted copy of an array.
histogram
Produce histogram from 1-D data.

Notes
Binary search is used to find the required insertion points.
As of Numpy 1.4.0 searchsorted works with real/complex arrays containing nan values. The enhanced
sort order is documented in sort.

Examples
>>> np.searchsorted([1,2,3,4,5], 3)
2
>>> np.searchsorted([1,2,3,4,5], 3, side=’right’)
3
>>> np.searchsorted([1,2,3,4,5], [-10, 10, 2, 3])
array([0, 5, 1, 2])

numpy.sort(a, axis=-1, kind=’quicksort’, order=None)

ndarray.sort
Method to sort an array in-place.

210 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

argsort
Indirect sort.
lexsort
Indirect stable sort on multiple keys.
searchsorted
Find elements in a sorted array.

Sort by age, then height if ages are equal:

1.5. Standard array subclasses 211

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.sort(a, order=[’age’, ’height’])

array([(’Galahad’, 1.7, 38), (’Lancelot’, 1.8999999999999999, 38),
(’Arthur’, 1.8, 41)],
dtype=[(’name’, ’|S10’), (’height’, ’<f8’), (’age’, ’<i4’)])

Examples
>>> x = np.array([[[0], [1], [2]]])
>>> x.shape
(1, 3, 1)
>>> np.squeeze(x).shape
(3,)

numpy.std(a, axis=None, dtype=None, out=None, ddof=0)

212 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:
var, mean
numpy.doc.ufuncs
Section “Output arguments”

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.std(a)
1.1180339887498949
>>> np.std(a, axis=0)
array([ 1., 1.])
>>> np.std(a, axis=1)
array([ 0.5, 0.5])

In single precision, std() can be inaccurate:

>>> a = np.zeros((2,512*512), dtype=np.float32)
>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.std(a)
0.45172946707416706

Computing the standard deviation in float64 is more accurate:

>>> np.std(a, dtype=np.float64)
0.44999999925552653

numpy.sum(a, axis=None, dtype=None, out=None)

1.5. Standard array subclasses 213

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.sum
Equivalent method.
cumsum
Cumulative sum of array elements.
trapz
Integration of array values using the composite trapezoidal rule.

mean, average

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

If the accumulator is too small, overflow occurs:

>>> np.ones(128, dtype=np.int8).sum(dtype=np.int8)
-128

numpy.swapaxes(a, axis1, axis2)

Interchange two axes of an array.
Parameters
a : array_like
Input array.

214 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

axis1 : int
First axis.
axis2 : int
Second axis.
Returns
a_swapped : ndarray
If a is an ndarray, then a view of a is returned; otherwise a new array is created.

Examples
>>> x = np.array([[1,2,3]])
>>> np.swapaxes(x,0,1)
array([[1],
[2],
[3]])

>>> x = np.array([[[0,1],[2,3]],[[4,5],[6,7]]])
>>> x
array([[[0, 1],
[2, 3]],
[[4, 5],
[6, 7]]])

>>> np.swapaxes(x,0,2)
array([[[0, 4],
[2, 6]],
[[1, 5],
[3, 7]]])

numpy.take(a, indices, axis=None, out=None, mode=’raise’)

1.5. Standard array subclasses 215

NumPy Reference, Release 2.0.0.dev-ac3cba3

• ‘clip’ – clip to the range

ndarray.take
equivalent method

Examples
>>> a = [4, 3, 5, 7, 6, 8]
>>> indices = [0, 1, 4]
>>> np.take(a, indices)
array([4, 3, 6])

In this example if a is an ndarray, “fancy” indexing can be used.

>>> a = np.array(a)
>>> a[indices]
array([4, 3, 6])

numpy.trace(a, offset=0, axis1=0, axis2=1, dtype=None, out=None)

216 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> np.trace(np.eye(3))
3.0
>>> a = np.arange(8).reshape((2,2,2))
>>> np.trace(a)
array([6, 8])

>>> a = np.arange(24).reshape((2,2,2,3))
>>> np.trace(a).shape
(2, 3)

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.transpose(x)
array([[0, 2],
[1, 3]])

>>> x = np.ones((1, 2, 3))

>>> np.transpose(x, (1, 0, 2)).shape
(2, 1, 3)

1.5. Standard array subclasses 217

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.var(a, axis=None, dtype=None, out=None, ddof=0)

std
Standard deviation
mean
Average
numpy.doc.ufuncs
Section “Output arguments”

218 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> a = np.array([[1,2],[3,4]])
>>> np.var(a)
1.25
>>> np.var(a,0)
array([ 1., 1.])
>>> np.var(a,1)
array([ 0.25, 0.25])

In single precision, var() can be inaccurate:

>>> a = np.zeros((2,512*512), dtype=np.float32)
>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.var(a)
0.20405951142311096

Computing the standard deviation in float64 is more accurate:

>>> np.var(a, dtype=np.float64)
0.20249999932997387
>>> ((1-0.55)**2 + (0.1-0.55)**2)/2
0.20250000000000001

class numpy.record
A data-type scalar that allows field access as attribute lookup.

Methods

1.5. Standard array subclasses 219

NumPy Reference, Release 2.0.0.dev-ac3cba3

Table 1.6 – continued from previous page

itemset
max(a[, axis, out]) Return the maximum of an array or maximum along an axis.
mean(a[, axis, dtype, out]) Compute the arithmetic mean along the specified axis.
min(a[, axis, out]) Return the minimum of an array or minimum along an axis.
newbyteorder
nonzero(a) Return the indices of the elements that are non-zero.
pprint Support to pretty-print lists, tuples, & dictionaries recursively.
prod(a[, axis, dtype, out]) Return the product of array elements over a given axis.
ptp(a[, axis, out]) Range of values (maximum - minimum) along an axis.
put(a, ind, v[, mode]) Replaces specified elements of an array with given values.
ravel(a[, order]) Return a flattened array.
repeat(a, repeats[, axis]) Repeat elements of an array.
reshape(a, newshape[, order]) Gives a new shape to an array without changing its data.
resize(a, new_shape) Return a new array with the specified shape.
round(a[, decimals, out]) Round an array to the given number of decimals.
searchsorted(a, v[, side]) Find indices where elements should be inserted to maintain order.
setfield
setflags
sort(a[, axis, kind, order]) Return a sorted copy of an array.
squeeze(a) Remove single-dimensional entries from the shape of an array.
std(a[, axis, dtype, out, ddof]) Compute the standard deviation along the specified axis.
sum(a[, axis, dtype, out]) Sum of array elements over a given axis.
swapaxes(a, axis1, axis2) Interchange two axes of an array.
take(a, indices[, axis, out, mode]) Take elements from an array along an axis.
tofile
tolist
tostring
trace(a[, offset, axis1, axis2, dtype, out]) Return the sum along diagonals of the array.
transpose(a[, axes]) Permute the dimensions of an array.
var(a[, axis, dtype, out, ddof]) Compute the variance along the specified axis.
view

numpy.all(a, axis=None, out=None)

220 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

A new boolean or array is returned unless out is specified, in which case a reference to
out is returned.
See Also:

ndarray.all
equivalent method
any
Test whether any element along a given axis evaluates to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.all([[True,False],[True,True]])
False

>>> np.all([[True,False],[True,True]], axis=0)

array([ True, False], dtype=bool)

>>> np.all([-1, 4, 5])

True

>>> np.all([1.0, np.nan])

True

>>> o=np.array([False])
>>> z=np.all([-1, 4, 5], out=o)
>>> id(z), id(o), z
(28293632, 28293632, array([ True], dtype=bool))

numpy.any(a, axis=None, out=None)

1.5. Standard array subclasses 221

NumPy Reference, Release 2.0.0.dev-ac3cba3

A new boolean or ndarray is returned unless out is specified, in which case a reference
to out is returned.
See Also:

ndarray.any
equivalent method
all
Test whether all elements along a given axis evaluate to True.

Notes
Not a Number (NaN), positive infinity and negative infinity evaluate to True because these are not equal to
zero.

Examples
>>> np.any([[True, False], [True, True]])
True

>>> np.any([[True, False], [False, False]], axis=0)

array([ True, False], dtype=bool)

>>> np.any([-1, 0, 5])

True

>>> np.any(np.nan)
True

222 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

amax
The maximum value along a given axis.
unravel_index
Convert a flat index into an index tuple.

Notes
In case of multiple occurrences of the maximum values, the indices corresponding to the first occurrence
are returned.

Examples
>>> a = np.arange(6).reshape(2,3)
>>> a
array([[0, 1, 2],
[3, 4, 5]])
>>> np.argmax(a)
5
>>> np.argmax(a, axis=0)
array([1, 1, 1])
>>> np.argmax(a, axis=1)
array([2, 2])

>>> b = np.arange(6)
>>> b[1] = 5
>>> b
array([0, 5, 2, 3, 4, 5])
>>> np.argmax(b) # Only the first occurrence is returned.
1

numpy.argmin(a, axis=None)
Return the indices of the minimum values along an axis.
See Also:

argmax
Similar function. Please refer to numpy.argmax for detailed documentation.

numpy.argsort(a, axis=-1, kind=’quicksort’, order=None)

1.5. Standard array subclasses 223

NumPy Reference, Release 2.0.0.dev-ac3cba3

When a is an array with fields defined, this argument specifies which fields to compare
first, second, etc. Not all fields need be specified.
Returns
index_array : ndarray, int
Array of indices that sort a along the specified axis. In other words,
a[index_array] yields a sorted a.
See Also:

sort
Describes sorting algorithms used.
lexsort
Indirect stable sort with multiple keys.
ndarray.sort
Inplace sort.

Notes
See sort for notes on the different sorting algorithms.
As of NumPy 1.4.0 argsort works with real/complex arrays containing nan values. The enhanced sort
order is documented in sort.

Examples
One dimensional array:
>>> x = np.array([3, 1, 2])
>>> np.argsort(x)
array([1, 2, 0])

Two-dimensional array:
>>> x = np.array([[0, 3], [2, 2]])
>>> x
array([[0, 3],
[2, 2]])

>>> np.argsort(x, axis=0)

array([[0, 1],
[1, 0]])

>>> np.argsort(x, axis=1)

array([[0, 1],
[0, 1]])

Sorting with keys:

>>> x = np.array([(1, 0), (0, 1)], dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])
>>> x
array([(1, 0), (0, 1)],
dtype=[(’x’, ’<i4’), (’y’, ’<i4’)])

>>> np.argsort(x, order=(’x’,’y’))

array([1, 0])

224 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.argsort(x, order=(’y’,’x’))

array([0, 1])

numpy.choose(a, choices, out=None, mode=’raise’)

1.5. Standard array subclasses 225

NumPy Reference, Release 2.0.0.dev-ac3cba3

If a and each choice array are not all broadcastable to the same shape.

See Also:

ndarray.choose
equivalent method

A couple examples illustrating how choose broadcasts:

>>> a = [[1, 0, 1], [0, 1, 0], [1, 0, 1]]
>>> choices = [-10, 10]
>>> np.choose(a, choices)
array([[ 10, -10, 10],
[-10, 10, -10],
[ 10, -10, 10]])

>>> # With thanks to Anne Archibald

numpy.clip(a, a_min, a_max, out=None)

226 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Array containing elements to clip.

a_min : scalar or array_like
Minimum value.
a_max : scalar or array_like
Maximum value. If a_min or a_max are array_like, then they will be broadcasted to the
shape of a.
out : ndarray, optional
The results will be placed in this array. It may be the input array for in-place clipping.
out must be of the right shape to hold the output. Its type is preserved.
Returns
clipped_array : ndarray
An array with the elements of a, but where values < a_min are replaced with a_min, and
those > a_max with a_max.
See Also:

numpy.doc.ufuncs
Section “Output arguments”

numpy.compress(condition, a, axis=None, out=None)

1.5. Standard array subclasses 227

NumPy Reference, Release 2.0.0.dev-ac3cba3

Output array. Its type is preserved and it must be of the right shape to hold the output.
Returns
compressed_array : ndarray
A copy of a without the slices along axis for which condition is false.
See Also:
take, choose, diag, diagonal, select
ndarray.compress
Equivalent method.
numpy.doc.ufuncs
Section “Output arguments”

Working on the flattened array does not return slices along an axis but selects elements.
>>> np.compress([False, True], a)
array([2])

Examples
>>> np.conjugate(1+2j)
(1-2j)

>>> x = np.eye(2) + 1j * np.eye(2)

>>> np.conjugate(x)
array([[ 1.-1.j, 0.-0.j],
[ 0.-0.j, 1.-1.j]])

228 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.copy(a)
Return an array copy of the given object.
Parameters
a : array_like
Input data.
Returns
arr : ndarray
Array interpretation of a.

Notes
This is equivalent to
>>> np.array(a, copy=True)

Examples
Create an array x, with a reference y and a copy z:
>>> x = np.array([1, 2, 3])
>>> y = x
>>> z = np.copy(x)

Note that, when we modify x, y changes, but not z:

>>> x[0] = 10
>>> x[0] == y[0]
True
>>> x[0] == z[0]
False

numpy.cumprod(a, axis=None, dtype=None, out=None)

1.5. Standard array subclasses 229

NumPy Reference, Release 2.0.0.dev-ac3cba3

A new array holding the result is returned unless out is specified, in which case a refer-
ence to out is returned.
See Also:

numpy.doc.ufuncs
Section “Output arguments”

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

The cumulative product for each column (i.e., over the rows) of a:
>>> np.cumprod(a, axis=0)
array([[ 1, 2, 3],
[ 4, 10, 18]])

The cumulative product for each row (i.e. over the columns) of a:
>>> np.cumprod(a,axis=1)
array([[ 1, 2, 6],
[ 4, 20, 120]])

numpy.cumsum(a, axis=None, dtype=None, out=None)

230 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

A new array holding the result is returned unless out is specified, in which case a ref-
erence to out is returned. The result has the same size as a, and the same shape as a if
axis is not None or a is a 1-d array.
See Also:

sum
Sum array elements.
trapz
Integration of array values using the composite trapezoidal rule.

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

>>> np.cumsum(a,axis=0) # sum over rows for each of the 3 columns

array([[1, 2, 3],
[5, 7, 9]])
>>> np.cumsum(a,axis=1) # sum over columns for each of the 2 rows
array([[ 1, 3, 6],
[ 4, 9, 15]])

numpy.diagonal(a, offset=0, axis1=0, axis2=1)

1.5. Standard array subclasses 231

NumPy Reference, Release 2.0.0.dev-ac3cba3

Returns
array_of_diagonals : ndarray
If a is 2-D, a 1-D array containing the diagonal is returned. If the dimension of a is
larger, then an array of diagonals is returned, “packed” from left-most dimension to
right-most (e.g., if a is 3-D, then the diagonals are “packed” along rows).
Raises
ValueError :
If the dimension of a is less than 2.
See Also:

diag
MATLAB work-a-like for 1-D and 2-D arrays.
diagflat
Create diagonal arrays.
trace
Sum along diagonals.

Examples
>>> a = np.arange(4).reshape(2,2)
>>> a
array([[0, 1],
[2, 3]])
>>> a.diagonal()
array([0, 3])
>>> a.diagonal(1)
array([1])

numpy.mean(a, axis=None, dtype=None, out=None)

Compute the arithmetic mean along the specified axis.

232 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Returns the average of the array elements. The average is taken over the flattened array by default, other-
wise over the specified axis. float64 intermediate and return values are used for integer inputs.
Parameters
a : array_like
Array containing numbers whose mean is desired. If a is not an array, a conversion is
attempted.
axis : int, optional
Axis along which the means are computed. The default is to compute the mean of the
flattened array.
dtype : data-type, optional
Type to use in computing the mean. For integer inputs, the default is float64; for floating
point inputs, it is the same as the input dtype.
out : ndarray, optional
Alternate output array in which to place the result. The default is None; if provided, it
must have the same shape as the expected output, but the type will be cast if necessary.
See doc.ufuncs for details.
Returns
m : ndarray, see dtype parameter above
If out=None, returns a new array containing the mean values, otherwise a reference to
the output array is returned.
See Also:

average
Weighted average

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.mean(a)
2.5
>>> np.mean(a, axis=0)
array([ 2., 3.])
>>> np.mean(a, axis=1)
array([ 1.5, 3.5])

In single precision, mean can be inaccurate:

>>> a = np.zeros((2, 512*512), dtype=np.float32)
>>> a[0, :] = 1.0
>>> a[1, :] = 0.1
>>> np.mean(a)
0.546875

1.5. Standard array subclasses 233

NumPy Reference, Release 2.0.0.dev-ac3cba3

Computing the mean in float64 is more accurate:

>>> np.mean(a, dtype=np.float64)
0.55000000074505806

To group the indices by element, rather than dimension, use:

transpose(nonzero(a))

The result of this is always a 2-D array, with a row for each non-zero element.
Parameters
a : array_like
Input array.
Returns
tuple_of_arrays : tuple
Indices of elements that are non-zero.
See Also:

Examples
>>> x = np.eye(3)
>>> x
array([[ 1., 0., 0.],
[ 0., 1., 0.],
[ 0., 0., 1.]])
>>> np.nonzero(x)
(array([0, 1, 2]), array([0, 1, 2]))

>>> x[np.nonzero(x)]
array([ 1., 1., 1.])
>>> np.transpose(np.nonzero(x))
array([[0, 0],
[1, 1],
[2, 2]])

234 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

The nonzero method of the boolean array can also be called.

>>> (a > 3).nonzero()
(array([1, 1, 1, 2, 2, 2]), array([0, 1, 2, 0, 1, 2]))

numpy.prod(a, axis=None, dtype=None, out=None)

ndarray.prod
equivalent method
numpy.doc.ufuncs
Section “Output arguments”

1.5. Standard array subclasses 235

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
By default, calculate the product of all elements:
>>> np.prod([1.,2.])
2.0

Even when the input array is two-dimensional:

>>> np.prod([[1.,2.],[3.,4.]])
24.0

But we can also specify the axis over which to multiply:

>>> np.prod([[1.,2.],[3.,4.]], axis=1)
array([ 2., 12.])

If the type of x is unsigned, then the output type is the unsigned platform integer:
>>> x = np.array([1, 2, 3], dtype=np.uint8)
>>> np.prod(x).dtype == np.uint
True

If x is of a signed integer type, then the output type is the default platform integer:
>>> x = np.array([1, 2, 3], dtype=np.int8)
>>> np.prod(x).dtype == np.int
True

numpy.ptp(a, axis=None, out=None)

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.ptp(x, axis=0)

array([2, 2])

236 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.ptp(x, axis=1)

array([1, 1])

numpy.put(a, ind, v, mode=’raise’)

Replaces specified elements of an array with given values.
The indexing works on the flattened target array. put is roughly equivalent to:
a.flat[ind] = v

See Also:
putmask, place

Examples
>>> a = np.arange(5)
>>> np.put(a, [0, 2], [-44, -55])
>>> a
array([-44, 1, -55, 3, 4])

>>> a = np.arange(5)
>>> np.put(a, 22, -5, mode=’clip’)
>>> a
array([ 0, 1, 2, 3, -5])

numpy.ravel(a, order=’C’)
Return a flattened array.
A 1-D array, containing the elements of the input, is returned. A copy is made only if needed.
Parameters
a : array_like

1.5. Standard array subclasses 237

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.flat
1-D iterator over an array.
ndarray.flatten
1-D array copy of the elements of an array in row-major order.

Examples
It is equivalent to reshape(-1, order=order).
>>> x = np.array([[1, 2, 3], [4, 5, 6]])
>>> print np.ravel(x)
[1 2 3 4 5 6]

>>> print x.reshape(-1)

[1 2 3 4 5 6]

>>> print np.ravel(x, order=’F’)

[1 4 2 5 3 6]

When order is ‘A’, it will preserve the array’s ‘C’ or ‘F’ ordering:
>>> print np.ravel(x.T)
[1 4 2 5 3 6]
>>> print np.ravel(x.T, order=’A’)
[1 2 3 4 5 6]

238 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.repeat(a, repeats, axis=None)

tile
Tile an array.

numpy.reshape(a, newshape, order=’C’)

Gives a new shape to an array without changing its data.
Parameters
a : array_like
Array to be reshaped.
newshape : int or tuple of ints

1.5. Standard array subclasses 239

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.reshape
Equivalent method.

Examples
>>> a = np.array([[1,2,3], [4,5,6]])
>>> np.reshape(a, 6)
array([1, 2, 3, 4, 5, 6])
>>> np.reshape(a, 6, order=’F’)
array([1, 4, 2, 5, 3, 6])

>>> np.reshape(a, (3,-1)) # the unspecified value is inferred to be 2

array([[1, 2],
[3, 4],
[5, 6]])

240 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

ndarray.resize
resize an array in-place.

Examples
>>> a=np.array([[0,1],[2,3]])
>>> np.resize(a,(1,4))
array([[0, 1, 2, 3]])
>>> np.resize(a,(2,4))
array([[0, 1, 2, 3],
[0, 1, 2, 3]])

sort
Return a sorted copy of an array.
histogram
Produce histogram from 1-D data.

Notes
Binary search is used to find the required insertion points.
As of Numpy 1.4.0 searchsorted works with real/complex arrays containing nan values. The enhanced
sort order is documented in sort.

1.5. Standard array subclasses 241

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> np.searchsorted([1,2,3,4,5], 3)
2
>>> np.searchsorted([1,2,3,4,5], 3, side=’right’)
3
>>> np.searchsorted([1,2,3,4,5], [-10, 10, 2, 3])
array([0, 5, 1, 2])

numpy.sort(a, axis=-1, kind=’quicksort’, order=None)

ndarray.sort
Method to sort an array in-place.
argsort
Indirect sort.
lexsort
Indirect stable sort on multiple keys.
searchsorted
Find elements in a sorted array.

242 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Sort by age, then height if ages are equal:

1.5. Standard array subclasses 243

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> x = np.array([[[0], [1], [2]]])
>>> x.shape
(1, 3, 1)
>>> np.squeeze(x).shape
(3,)

numpy.std(a, axis=None, dtype=None, out=None, ddof=0)

244 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> np.std(a)
1.1180339887498949
>>> np.std(a, axis=0)
array([ 1., 1.])
>>> np.std(a, axis=1)
array([ 0.5, 0.5])

In single precision, std() can be inaccurate:

>>> a = np.zeros((2,512*512), dtype=np.float32)
>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.std(a)
0.45172946707416706

Computing the standard deviation in float64 is more accurate:

>>> np.std(a, dtype=np.float64)
0.44999999925552653

numpy.sum(a, axis=None, dtype=None, out=None)

1.5. Standard array subclasses 245

NumPy Reference, Release 2.0.0.dev-ac3cba3

See Also:

ndarray.sum
Equivalent method.
cumsum
Cumulative sum of array elements.
trapz
Integration of array values using the composite trapezoidal rule.

mean, average

Notes
Arithmetic is modular when using integer types, and no error is raised on overflow.

If the accumulator is too small, overflow occurs:

>>> np.ones(128, dtype=np.int8).sum(dtype=np.int8)
-128

numpy.swapaxes(a, axis1, axis2)

Examples
>>> x = np.array([[1,2,3]])
>>> np.swapaxes(x,0,1)
array([[1],
[2],
[3]])

246 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x = np.array([[[0,1],[2,3]],[[4,5],[6,7]]])
>>> x
array([[[0, 1],
[2, 3]],
[[4, 5],
[6, 7]]])

>>> np.swapaxes(x,0,2)
array([[[0, 4],
[2, 6]],
[[1, 5],
[3, 7]]])

numpy.take(a, indices, axis=None, out=None, mode=’raise’)

ndarray.take
equivalent method

1.5. Standard array subclasses 247

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> a = [4, 3, 5, 7, 6, 8]
>>> indices = [0, 1, 4]
>>> np.take(a, indices)
array([4, 3, 6])

In this example if a is an ndarray, “fancy” indexing can be used.

>>> a = np.array(a)
>>> a[indices]
array([4, 3, 6])

numpy.trace(a, offset=0, axis1=0, axis2=1, dtype=None, out=None)

Examples

248 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> np.trace(np.eye(3))
3.0
>>> a = np.arange(8).reshape((2,2,2))
>>> np.trace(a)
array([6, 8])

>>> a = np.arange(24).reshape((2,2,2,3))
>>> np.trace(a).shape
(2, 3)

Examples
>>> x = np.arange(4).reshape((2,2))
>>> x
array([[0, 1],
[2, 3]])

>>> np.transpose(x)
array([[0, 2],
[1, 3]])

>>> x = np.ones((1, 2, 3))

>>> np.transpose(x, (1, 0, 2)).shape
(2, 1, 3)

numpy.var(a, axis=None, dtype=None, out=None, ddof=0)

1.5. Standard array subclasses 249

NumPy Reference, Release 2.0.0.dev-ac3cba3

dtype : data-type, optional

std
Standard deviation
mean
Average
numpy.doc.ufuncs
Section “Output arguments”

Examples
>>> a = np.array([[1,2],[3,4]])
>>> np.var(a)
1.25
>>> np.var(a,0)
array([ 1., 1.])
>>> np.var(a,1)
array([ 0.25, 0.25])

In single precision, var() can be inaccurate:

250 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> a = np.zeros((2,512*512), dtype=np.float32)

>>> a[0,:] = 1.0
>>> a[1,:] = 0.1
>>> np.var(a)
0.20405951142311096

Computing the standard deviation in float64 is more accurate:

>>> np.var(a, dtype=np.float64)
0.20249999932997387
>>> ((1-0.55)**2 + (0.1-0.55)**2)/2
0.20250000000000001

1.5.6 Masked arrays (numpy.ma)

1.5.7 Standard container class

For backward compatibility and as a standard “container “class, the UserArray from Numeric has been brought over to
NumPy and named numpy.lib.user_array.container The container class is a Python class whose self.array
attribute is an ndarray. Multiple inheritance is probably easier with numpy.lib.user_array.container than with the
ndarray itself and so it is included by default. It is not documented here beyond mentioning its existence because you
are encouraged to use the ndarray class directly if you can.
numpy.lib.user_array.container(data[, ...])
class numpy.lib.user_array.container(data, dtype=None, copy=True)

1.5.8 Array Iterators

Iterators are a powerful concept for array processing. Essentially, iterators implement a generalized for-loop. If myiter
is an iterator object, then the Python code:
for val in myiter:
...
some code involving val
...

calls val = myiter.next() repeatedly until StopIteration is raised by the iterator. There are several ways
to iterate over an array that may be useful: default iteration, flat iteration, and N -dimensional enumeration.

Default iteration

The default iterator of an ndarray object is the default Python iterator of a sequence type. Thus, when the array object
itself is used as an iterator. The default behavior is equivalent to:
for i in xrange(arr.shape[0]):
val = arr[i]

1.5. Standard array subclasses 251

NumPy Reference, Release 2.0.0.dev-ac3cba3

This default iterator selects a sub-array of dimension N − 1 from the array. This can be a useful construct for defining
recursive algorithms. To loop over the entire array requires N for-loops.
>>> a = arange(24).reshape(3,2,4)+10
>>> for val in a:
... print ’item:’, val
item: [[10 11 12 13]
[14 15 16 17]]
item: [[18 19 20 21]
[22 23 24 25]]
item: [[26 27 28 29]
[30 31 32 33]]

Flat iteration

ndarray.flat A 1-D iterator over the array.

ndarray.flat
A 1-D iterator over the array.
This is a numpy.flatiter instance, which acts similarly to, but is not a subclass of, Python’s built-in iterator
object.
See Also:

flatten
Return a copy of the array collapsed into one dimension.

flatiter

Examples
>>> x = np.arange(1, 7).reshape(2, 3)
>>> x
array([[1, 2, 3],
[4, 5, 6]])
>>> x.flat[3]
4
>>> x.T
array([[1, 4],
[2, 5],
[3, 6]])
>>> x.T.flat[3]
5
>>> type(x.flat)
<type ’numpy.flatiter’>

An assignment example:
>>> x.flat = 3; x
array([[3, 3, 3],
[3, 3, 3]])
>>> x.flat[[1,4]] = 1; x
array([[3, 1, 3],
[3, 1, 3]])

As mentioned previously, the flat attribute of ndarray objects returns an iterator that will cycle over the entire array in
C-style contiguous order.

252 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> for i, val in enumerate(a.flat):

... if i%5 == 0: print i, val
0 10
5 15
10 20
15 25
20 30

Here, I’ve used the built-in enumerate iterator to return the iterator index as well as the value.

N-dimensional enumeration

ndenumerate(arr) Multidimensional index iterator.

class numpy.ndenumerate(arr)
Multidimensional index iterator.
Return an iterator yielding pairs of array coordinates and values.
Parameters
a : ndarray
Input array.
See Also:
ndindex, flatiter

Examples
>>> a = np.array([[1, 2], [3, 4]])
>>> for index, x in np.ndenumerate(a):
... print index, x
(0, 0) 1
(0, 1) 2
(1, 0) 3
(1, 1) 4

Methods
next
Sometimes it may be useful to get the N-dimensional index while iterating. The ndenumerate iterator can achieve this.
>>> for i, val in ndenumerate(a):
... if sum(i)%5 == 0: print i, val
(0, 0, 0) 10
(1, 1, 3) 25
(2, 0, 3) 29
(2, 1, 2) 32

Iterator for broadcasting

broadcast Produce an object that mimics broadcasting.

class numpy.broadcast
Produce an object that mimics broadcasting.

1.5. Standard array subclasses 253

NumPy Reference, Release 2.0.0.dev-ac3cba3

Parameters
in1, in2, ... : array_like
Input parameters.
Returns
b : broadcast object
Broadcast the input parameters against one another, and return an object that encapsu-
lates the result. Amongst others, it has shape and nd properties, and may be used as
an iterator.

Examples
Manually adding two vectors, using broadcasting:
>>> x = np.array([[1], [2], [3]])
>>> y = np.array([4, 5, 6])
>>> b = np.broadcast(x, y)

>>> out = np.empty(b.shape)

>>> out.flat = [u+v for (u,v) in b]
>>> out
array([[ 5., 6., 7.],
[ 6., 7., 8.],
[ 7., 8., 9.]])

Compare against built-in broadcasting:

>>> x + y
array([[5, 6, 7],
[6, 7, 8],
[7, 8, 9]])

Methods
next
reset
The general concept of broadcasting is also available from Python using the broadcast iterator. This object takes
N objects as inputs and returns an iterator that returns tuples providing each of the input sequence elements in the
broadcasted result.
>>> for val in broadcast([[1,0],[2,3]],[0,1]):
... print val
(1, 0)
(0, 1)
(2, 0)
(3, 1)

1.6 Masked arrays

Masked arrays are arrays that may have missing or invalid entries. The numpy.ma module provides a nearly work-
alike replacement for numpy that supports data arrays with masks.

254 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

1.6.1 The numpy.ma module

Rationale

Masked arrays are arrays that may have missing or invalid entries. The numpy.ma module provides a nearly work-
alike replacement for numpy that supports data arrays with masks.

What is a masked array?

In many circumstances, datasets can be incomplete or tainted by the presence of invalid data. For example, a sensor
may have failed to record a data, or recorded an invalid value. The numpy.ma module provides a convenient way to
address this issue, by introducing masked arrays.
A masked array is the combination of a standard numpy.ndarray and a mask. A mask is either nomask, indicating
that no value of the associated array is invalid, or an array of booleans that determines for each element of the associated
array whether the value is valid or not. When an element of the mask is False, the corresponding element of the
associated array is valid and is said to be unmasked. When an element of the mask is True, the corresponding element
of the associated array is said to be masked (invalid).
The package ensures that masked entries are not used in computations.
As an illustration, let’s consider the following dataset:
>>> import numpy as np
>>> import numpy.ma as ma
>>> x = np.array([1, 2, 3, -1, 5])

We wish to mark the fourth entry as invalid. The easiest is to create a masked array:
>>> mx = ma.masked_array(x, mask=[0, 0, 0, 1, 0])

We can now compute the mean of the dataset, without taking the invalid data into account:
>>> mx.mean()
2.75

The numpy.ma module

The main feature of the numpy.ma module is the MaskedArray class, which is a subclass of numpy.ndarray.
The class, its attributes and methods are described in more details in the MaskedArray class section.
The numpy.ma module can be used as an addition to numpy:
>>> import numpy as np
>>> import numpy.ma as ma

To create an array with the second element invalid, we would do:

>>> y = ma.array([1, 2, 3], mask = [0, 1, 0])

To create a masked array where all values close to 1.e20 are invalid, we would do:
>>> z = masked_values([1.0, 1.e20, 3.0, 4.0], 1.e20)

For a complete discussion of creation methods for masked arrays please see section Constructing masked arrays.

1.6. Masked arrays 255

NumPy Reference, Release 2.0.0.dev-ac3cba3

1.6.2 Using numpy.ma

Constructing masked arrays

There are several ways to construct a masked array.

• A first possibility is to directly invoke the MaskedArray class.
• A second possibility is to use the two masked array constructors, array and masked_array.
array(data[, dtype, copy, order, mask, ...]) An array class with possibly masked values.
masked_array An array class with possibly masked values.
numpy.ma.array(data, dtype=None, copy=False, order=False, mask=False, fill_value=None,
keep_mask=True, hard_mask=False, shrink=True, subok=True, ndmin=0)
An array class with possibly masked values.
Masked values of True exclude the corresponding element from any computation.
Construction:
x = MaskedArray(data, mask=nomask, dtype=None,
copy=False, subok=True, ndmin=0, fill_value=None,
keep_mask=True, hard_mask=None, shrink=True)

Parameters
data : array_like
Input data.
mask : sequence, optional
Mask. Must be convertible to an array of booleans with the same shape as data. True
indicates a masked (i.e. invalid) data.
dtype : dtype, optional
Data type of the output. If dtype is None, the type of the data argument (data.dtype)
is used. If dtype is not None and different from data.dtype, a copy is performed.
copy : bool, optional
Whether to copy the input data (True), or to use a reference instead. Default is False.
subok : bool, optional
Whether to return a subclass of MaskedArray if possible (True) or a plain MaskedArray.
Default is True.
ndmin : int, optional
Minimum number of dimensions. Default is 0.
fill_value : scalar, optional
Value used to fill in the masked values when necessary. If None, a default based on the
data-type is used.
keep_mask : bool, optional
Whether to combine mask with the mask of the input data, if any (True), or to use only
mask for the output (False). Default is True.
hard_mask : bool, optional

256 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Whether to use a hard mask or not. With a hard mask, masked values cannot be un-
masked. Default is False.
shrink : bool, optional
Whether to force compression of an empty mask. Default is True.

numpy.ma.masked_array
alias of MaskedArray
• A third option is to take the view of an existing array. In that case, the mask of the view is set to nomask if the
array has no named fields, or an array of boolean with the same structure as the array otherwise.
>>> x = np.array([1, 2, 3])
>>> x.view(ma.MaskedArray)
masked_array(data = [1 2 3],
mask = False,
fill_value = 999999)
>>> x = np.array([(1, 1.), (2, 2.)], dtype=[(’a’,int), (’b’, float)])
>>> x.view(ma.MaskedArray)
masked_array(data = [(1, 1.0) (2, 2.0)],
mask = [(False, False) (False, False)],
fill_value = (999999, 1e+20),
dtype = [(’a’, ’<i4’), (’b’, ’<f8’)])

• Yet another possibility is to use any of the following functions:

asarray(a[, dtype, order]) Convert the input to a masked array of the given data-type.
asanyarray(a[, dtype]) Convert the input to a masked array, conserving subclasses.
fix_invalid(a[, mask, copy, fill_value]) Return input with invalid data masked and replaced by a fill
value.
masked_equal(x, value[, copy]) Mask an array where equal to a given value.
masked_greater(x, value[, copy]) Mask an array where greater than a given value.
masked_greater_equal(x, value[, Mask an array where greater than or equal to a given value.
copy])
masked_inside(x, v1, v2[, copy]) Mask an array inside a given interval.
masked_invalid(a[, copy]) Mask an array where invalid values occur (NaNs or infs).
masked_less(x, value[, copy]) Mask an array where less than a given value.
masked_less_equal(x, value[, copy]) Mask an array where less than or equal to a given value.
masked_not_equal(x, value[, copy]) Mask an array where not equal to a given value.
masked_object(x, value[, copy, shrink]) Mask the array x where the data are exactly equal to value.
masked_outside(x, v1, v2[, copy]) Mask an array outside a given interval.
masked_values(x, value[, rtol, atol, Mask using floating point equality.
copy, ...])
masked_where(condition, a[, copy]) Mask an array where a condition is met.
numpy.ma.asarray(a, dtype=None, order=None)
Convert the input to a masked array of the given data-type.
No copy is performed if the input is already an ndarray. If a is a subclass of MaskedArray, a base class
MaskedArray is returned.
Parameters
a : array_like
Input data, in any form that can be converted to a masked array. This includes lists, lists
of tuples, tuples, tuples of tuples, tuples of lists, ndarrays and masked arrays.
dtype : dtype, optional

1.6. Masked arrays 257

NumPy Reference, Release 2.0.0.dev-ac3cba3

By default, the data-type is inferred from the input data.

order : {‘C’, ‘F’}, optional
Whether to use row-major (‘C’) or column-major (‘FORTRAN’) memory representa-
tion. Default is ‘C’.
Returns
out : MaskedArray
Masked array interpretation of a.
See Also:

asanyarray
Similar to asarray, but conserves subclasses.

Examples
>>> x = np.arange(10.).reshape(2, 5)
>>> x
array([[ 0., 1., 2., 3., 4.],
[ 5., 6., 7., 8., 9.]])
>>> np.ma.asarray(x)
masked_array(data =
[[ 0. 1. 2. 3. 4.]
[ 5. 6. 7. 8. 9.]],
mask =
False,
fill_value = 1e+20)
>>> type(np.ma.asarray(x))
<class ’numpy.ma.core.MaskedArray’>

numpy.ma.asanyarray(a, dtype=None)
Convert the input to a masked array, conserving subclasses.
If a is a subclass of MaskedArray, its class is conserved. No copy is performed if the input is already an
ndarray.
Parameters
a : array_like
Input data, in any form that can be converted to an array.
dtype : dtype, optional
By default, the data-type is inferred from the input data.
order : {‘C’, ‘F’}, optional
Whether to use row-major (‘C’) or column-major (‘FORTRAN’) memory representa-
tion. Default is ‘C’.
Returns
out : MaskedArray
MaskedArray interpretation of a.
See Also:

asarray
Similar to asanyarray, but does not conserve subclass.

258 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> x = np.arange(10.).reshape(2, 5)
>>> x
array([[ 0., 1., 2., 3., 4.],
[ 5., 6., 7., 8., 9.]])
>>> np.ma.asanyarray(x)
masked_array(data =
[[ 0. 1. 2. 3. 4.]
[ 5. 6. 7. 8. 9.]],
mask =
False,
fill_value = 1e+20)
>>> type(np.ma.asanyarray(x))
<class ’numpy.ma.core.MaskedArray’>

numpy.ma.fix_invalid(a, mask=False, copy=True, fill_value=None)

Return input with invalid data masked and replaced by a fill value.
Invalid data means values of nan, inf, etc.
Parameters
a : array_like
Input array, a (subclass of) ndarray.
copy : bool, optional
Whether to use a copy of a (True) or to fix a in place (False). Default is True.
fill_value : scalar, optional
Value used for fixing invalid data. Default is None, in which case the a.fill_value
is used.
Returns
b : MaskedArray
The input array with invalid entries fixed.

Notes
A copy is performed by default.

Examples
>>> x = np.ma.array([1., -1, np.nan, np.inf], mask=[1] + [0]*3)
>>> x
masked_array(data = [-- -1.0 nan inf],
mask = [ True False False False],
fill_value = 1e+20)
>>> np.ma.fix_invalid(x)
masked_array(data = [-- -1.0 -- --],
mask = [ True False True True],
fill_value = 1e+20)

>>> fixed = np.ma.fix_invalid(x)

>>> fixed.data
array([ 1.00000000e+00, -1.00000000e+00, 1.00000000e+20,
1.00000000e+20])
>>> x.data
array([ 1., -1., NaN, Inf])

1.6. Masked arrays 259

NumPy Reference, Release 2.0.0.dev-ac3cba3

numpy.ma.masked_equal(x, value, copy=True)

Mask an array where equal to a given value.
This function is a shortcut to masked_where, with condition = (x == value). For floating point arrays,
consider using masked_values(x, value).
See Also:

masked_where
Mask where a condition is met.
masked_values
Mask using floating point equality.

Examples
>>> import numpy.ma as ma
>>> a = np.arange(4)
>>> a
array([0, 1, 2, 3])
>>> ma.masked_equal(a, 2)
masked_array(data = [0 1 -- 3],
mask = [False False True False],
fill_value=999999)

numpy.ma.masked_greater(x, value, copy=True)

Mask an array where greater than a given value.
This function is a shortcut to masked_where, with condition = (x > value).
See Also:

masked_where
Mask where a condition is met.

Examples
>>> import numpy.ma as ma
>>> a = np.arange(4)
>>> a
array([0, 1, 2, 3])
>>> ma.masked_greater(a, 2)
masked_array(data = [0 1 2 --],
mask = [False False False True],
fill_value=999999)

numpy.ma.masked_greater_equal(x, value, copy=True)

Mask an array where greater than or equal to a given value.
This function is a shortcut to masked_where, with condition = (x >= value).
See Also:

masked_where
Mask where a condition is met.

Examples

260 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> import numpy.ma as ma

>>> a = np.arange(4)
>>> a
array([0, 1, 2, 3])
>>> ma.masked_greater_equal(a, 2)
masked_array(data = [0 1 -- --],
mask = [False False True True],
fill_value=999999)

numpy.ma.masked_inside(x, v1, v2, copy=True)

Mask an array inside a given interval.
Shortcut to masked_where, where condition is True for x inside the interval [v1,v2] (v1 <= x <= v2).
The boundaries v1 and v2 can be given in either order.
See Also:

masked_where
Mask where a condition is met.

Notes
The array x is prefilled with its filling value.

Examples
>>> import numpy.ma as ma
>>> x = [0.31, 1.2, 0.01, 0.2, -0.4, -1.1]
>>> ma.masked_inside(x, -0.3, 0.3)
masked_array(data = [0.31 1.2 -- -- -0.4 -1.1],
mask = [False False True True False False],
fill_value=1e+20)

The order of v1 and v2 doesn’t matter.

>>> ma.masked_inside(x, 0.3, -0.3)
masked_array(data = [0.31 1.2 -- -- -0.4 -1.1],
mask = [False False True True False False],
fill_value=1e+20)

numpy.ma.masked_invalid(a, copy=True)
Mask an array where invalid values occur (NaNs or infs).
This function is a shortcut to masked_where, with condition = ~(np.isfinite(a)). Any pre-existing mask
is conserved. Only applies to arrays with a dtype where NaNs or infs make sense (i.e. floating point types),
but accepts any array_like object.
See Also:

masked_where
Mask where a condition is met.

Examples
>>> import numpy.ma as ma
>>> a = np.arange(5, dtype=np.float)
>>> a[2] = np.NaN
>>> a[3] = np.PINF
>>> a

1.6. Masked arrays 261

NumPy Reference, Release 2.0.0.dev-ac3cba3

array([ 0., 1., NaN, Inf, 4.])

>>> ma.masked_invalid(a)
masked_array(data = [0.0 1.0 -- -- 4.0],
mask = [False False True True False],
fill_value=1e+20)

numpy.ma.masked_less(x, value, copy=True)

Mask an array where less than a given value.
This function is a shortcut to masked_where, with condition = (x < value).
See Also:

masked_where
Mask where a condition is met.

Examples
>>> import numpy.ma as ma
>>> a = np.arange(4)
>>> a
array([0, 1, 2, 3])
>>> ma.masked_less(a, 2)
masked_array(data = [-- -- 2 3],
mask = [ True True False False],
fill_value=999999)

numpy.ma.masked_less_equal(x, value, copy=True)

Mask an array where less than or equal to a given value.
This function is a shortcut to masked_where, with condition = (x <= value).
See Also:

masked_where
Mask where a condition is met.

Examples
>>> import numpy.ma as ma
>>> a = np.arange(4)
>>> a
array([0, 1, 2, 3])
>>> ma.masked_less_equal(a, 2)
masked_array(data = [-- -- -- 3],
mask = [ True True True False],
fill_value=999999)

numpy.ma.masked_not_equal(x, value, copy=True)

Mask an array where not equal to a given value.
This function is a shortcut to masked_where, with condition = (x != value).
See Also:

masked_where
Mask where a condition is met.

262 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Examples
>>> import numpy.ma as ma
>>> a = np.arange(4)
>>> a
array([0, 1, 2, 3])
>>> ma.masked_not_equal(a, 2)
masked_array(data = [-- -- 2 --],
mask = [ True True False True],
fill_value=999999)

numpy.ma.masked_object(x, value, copy=True, shrink=True)

Mask the array x where the data are exactly equal to value.
This function is similar to masked_values, but only suitable for object arrays: for floating point, use
masked_values instead.
Parameters
x : array_like
Array to mask
value : object
Comparison value
copy : {True, False}, optional
Whether to return a copy of x.
shrink : {True, False}, optional
Whether to collapse a mask full of False to nomask
Returns
result : MaskedArray
The result of masking x where equal to value.
See Also:

masked_where
Mask where a condition is met.
masked_equal
Mask where equal to a given value (integers).
masked_values
Mask using floating point equality.

Examples
>>> import numpy.ma as ma
>>> food = np.array([’green_eggs’, ’ham’], dtype=object)
>>> # don’t eat spoiled food
>>> eat = ma.masked_object(food, ’green_eggs’)
>>> print eat
[-- ham]
>>> # plain ol‘ ham is boring
>>> fresh_food = np.array([’cheese’, ’ham’, ’pineapple’], dtype=object)
>>> eat = ma.masked_object(fresh_food, ’green_eggs’)
>>> print eat
[cheese ham pineapple]

1.6. Masked arrays 263

NumPy Reference, Release 2.0.0.dev-ac3cba3

Note that mask is set to nomask if possible.

>>> eat
masked_array(data = [cheese ham pineapple],
mask = False,
fill_value=?)

numpy.ma.masked_outside(x, v1, v2, copy=True)

Mask an array outside a given interval.
Shortcut to masked_where, where condition is True for x outside the interval [v1,v2] (x < v1)|(x > v2).
The boundaries v1 and v2 can be given in either order.
See Also:

masked_where
Mask where a condition is met.

Notes
The array x is prefilled with its filling value.

Examples
>>> import numpy.ma as ma
>>> x = [0.31, 1.2, 0.01, 0.2, -0.4, -1.1]
>>> ma.masked_outside(x, -0.3, 0.3)
masked_array(data = [-- -- 0.01 0.2 -- --],
mask = [ True True False False True True],
fill_value=1e+20)

The order of v1 and v2 doesn’t matter.

>>> ma.masked_outside(x, 0.3, -0.3)
masked_array(data = [-- -- 0.01 0.2 -- --],
mask = [ True True False False True True],
fill_value=1e+20)

numpy.ma.masked_values(x, value, rtol=1.0000000000000001e-05, atol=1e-08, copy=True,

shrink=True)
Mask using floating point equality.
Return a MaskedArray, masked where the data in array x are approximately equal to value, i.e. where the
following condition is True
(abs(x - value) <= atol+rtol*abs(value))
The fill_value is set to value and the mask is set to nomask if possible. For integers, consider using
masked_equal.
Parameters
x : array_like
Array to mask.
value : float
Masking value.
rtol : float, optional
Tolerance parameter.
atol : float, optional

264 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

Tolerance parameter (1e-8).

copy : bool, optional
Whether to return a copy of x.
shrink : bool, optional
Whether to collapse a mask full of False to nomask.
Returns
result : MaskedArray
The result of masking x where approximately equal to value.
See Also:

masked_where
Mask where a condition is met.
masked_equal
Mask where equal to a given value (integers).

Examples
>>> import numpy.ma as ma
>>> x = np.array([1, 1.1, 2, 1.1, 3])
>>> ma.masked_values(x, 1.1)
masked_array(data = [1.0 -- 2.0 -- 3.0],
mask = [False True False True False],
fill_value=1.1)

Note that mask is set to nomask if possible.

>>> ma.masked_values(x, 1.5)
masked_array(data = [ 1. 1.1 2. 1.1 3. ],
mask = False,
fill_value=1.5)

For integers, the fill value will be different in general to the result of masked_equal.
>>> x = np.arange(5)
>>> x
array([0, 1, 2, 3, 4])
>>> ma.masked_values(x, 2)
masked_array(data = [0 1 -- 3 4],
mask = [False False True False False],
fill_value=2)
>>> ma.masked_equal(x, 2)
masked_array(data = [0 1 -- 3 4],
mask = [False False True False False],
fill_value=999999)

numpy.ma.masked_where(condition, a, copy=True)
Mask an array where a condition is met.
Return a as an array masked where condition is True. Any masked values of a or condition are also masked
in the output.
Parameters
condition : array_like

1.6. Masked arrays 265

NumPy Reference, Release 2.0.0.dev-ac3cba3

Masking condition. When condition tests floating point values for equality, consider
using masked_values instead.
a : array_like
Array to mask.
copy : bool
If True (default) make a copy of a in the result. If False modify a in place and return a
view.
Returns
result : MaskedArray
The result of masking a where condition is True.
See Also:

masked_values
Mask using floating point equality.
masked_equal
Mask where equal to a given value.
masked_not_equal
Mask where not equal to a given value.
masked_less_equal
Mask where less than or equal to a given value.
masked_greater_equal
Mask where greater than or equal to a given value.
masked_less
Mask where less than a given value.
masked_greater
Mask where greater than a given value.
masked_inside
Mask inside a given interval.
masked_outside
Mask outside a given interval.
masked_invalid
Mask invalid values (NaNs or infs).

Examples
>>> import numpy.ma as ma
>>> a = np.arange(4)
>>> a
array([0, 1, 2, 3])
>>> ma.masked_where(a <= 2, a)
masked_array(data = [-- -- -- 3],
mask = [ True True True False],
fill_value=999999)

Mask array b conditional on a.

266 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> b = [’a’, ’b’, ’c’, ’d’]

>>> ma.masked_where(a == 2, b)
masked_array(data = [a b -- d],
mask = [False False True False],
fill_value=N/A)

Effect of the copy argument.

>>> c = ma.masked_where(a <= 2, a)
>>> c
masked_array(data = [-- -- -- 3],
mask = [ True True True False],
fill_value=999999)
>>> c[0] = 99
>>> c
masked_array(data = [99 -- -- 3],
mask = [False True True False],
fill_value=999999)
>>> a
array([0, 1, 2, 3])
>>> c = ma.masked_where(a <= 2, a, copy=False)
>>> c[0] = 99
>>> c
masked_array(data = [99 -- -- 3],
mask = [False True True False],
fill_value=999999)
>>> a
array([99, 1, 2, 3])

When condition or a contain masked values.

>>> a = np.arange(4)
>>> a = ma.masked_where(a == 2, a)
>>> a
masked_array(data = [0 1 -- 3],
mask = [False False True False],
fill_value=999999)
>>> b = np.arange(4)
>>> b = ma.masked_where(b == 0, b)
>>> b
masked_array(data = [-- 1 2 3],
mask = [ True False False False],
fill_value=999999)
>>> ma.masked_where(a == 3, b)
masked_array(data = [-- 1 -- --],
mask = [ True False True True],
fill_value=999999)

Accessing the data

The underlying data of a masked array can be accessed in several ways:

• through the data attribute. The output is a view of the array as a numpy.ndarray or one of its subclasses,
depending on the type of the underlying data at the masked array creation.
• through the __array__ method. The output is then a numpy.ndarray.
• by directly taking a view of the masked array as a numpy.ndarray or one of its subclass (which is actually
what using the data attribute does).

1.6. Masked arrays 267

NumPy Reference, Release 2.0.0.dev-ac3cba3

• by using the getdata function.

None of these methods is completely satisfactory if some entries have been marked as invalid. As a general rule,
where a representation of the array is required without any masked entries, it is recommended to fill the array with the
filled method.

Accessing the mask

The mask of a masked array is accessible through its mask attribute. We must keep in mind that a True entry in the
mask indicates an invalid data.
Another possibility is to use the getmask and getmaskarray functions. getmask(x) outputs the mask of x if
x is a masked array, and the special value nomask otherwise. getmaskarray(x) outputs the mask of x if x is a
masked array. If x has no invalid entry or is not a masked array, the function outputs a boolean array of False with
as many elements as x.

Accessing only the valid entries

To retrieve only the valid entries, we can use the inverse of the mask as an index. The inverse of the mask can be
calculated with the numpy.logical_not function or simply with the ~ operator:
>>> x = ma.array([[1, 2], [3, 4]], mask=[[0, 1], [1, 0]])
>>> x[~x.mask]
masked_array(data = [1 4],
mask = [False False],
fill_value = 999999)

Another way to retrieve the valid data is to use the compressed method, which returns a one-dimensional ndarray
(or one of its subclasses, depending on the value of the baseclass attribute):
>>> x.compressed()
array([1, 4])

Note that the output of compressed is always 1D.

Modifying the mask

Masking an entry
The recommended way to mark one or several specific entries of a masked array as invalid is to assign the special
value masked to them:
>>> x = ma.array([1, 2, 3])
>>> x[0] = ma.masked
>>> x
masked_array(data = [-- 2 3],
mask = [ True False False],
fill_value = 999999)
>>> y = ma.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
>>> y[(0, 1, 2), (1, 2, 0)] = ma.masked
>>> y
masked_array(data =
[[1 -- 3]
[4 5 --]
[-- 8 9]],
mask =

268 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

[[False True False]

[False False True]
[ True False False]],
fill_value = 999999)
>>> z = ma.array([1, 2, 3, 4])
>>> z[:-2] = ma.masked
>>> z
masked_array(data = [-- -- 3 4],
mask = [ True True False False],
fill_value = 999999)

A second possibility is to modify the mask directly, but this usage is discouraged.
Note: When creating a new masked array with a simple, non-structured datatype, the mask is initially set to the
special value nomask, that corresponds roughly to the boolean False. Trying to set an element of nomask will fail
with a TypeError exception, as a boolean does not support item assignment.
All the entries of an array can be masked at once by assigning True to the mask:
>>> x = ma.array([1, 2, 3], mask=[0, 0, 1])
>>> x.mask = True
>>> x
masked_array(data = [-- -- --],
mask = [ True True True],
fill_value = 999999)

Finally, specific entries can be masked and/or unmasked by assigning to the mask a sequence of booleans:
>>> x = ma.array([1, 2, 3])
>>> x.mask = [0, 1, 0]
>>> x
masked_array(data = [1 -- 3],
mask = [False True False],
fill_value = 999999)

Unmasking an entry
To unmask one or several specific entries, we can just assign one or several new valid values to them:
>>> x = ma.array([1, 2, 3], mask=[0, 0, 1])
>>> x
masked_array(data = [1 2 --],
mask = [False False True],
fill_value = 999999)
>>> x[-1] = 5
>>> x
masked_array(data = [1 2 5],
mask = [False False False],
fill_value = 999999)

Note: Unmasking an entry by direct assignment will silently fail if the masked array has a hard mask, as shown by
the hardmask attribute. This feature was introduced to prevent overwriting the mask. To force the unmasking of an
entry where the array has a hard mask, the mask must first to be softened using the soften_mask method before the
allocation. It can be re-hardened with harden_mask:
>>> x = ma.array([1, 2, 3], mask=[0, 0, 1], hard_mask=True)
>>> x
masked_array(data = [1 2 --],
mask = [False False True],
fill_value = 999999)

1.6. Masked arrays 269

NumPy Reference, Release 2.0.0.dev-ac3cba3

>>> x[-1] = 5
>>> x
masked_array(data = [1 2 --],
mask = [False False True],
fill_value = 999999)
>>> x.soften_mask()
>>> x[-1] = 5
>>> x
masked_array(data = [1 2 5],
mask = [False False False],
fill_value = 999999)
>>> x.harden_mask()

To unmask all masked entries of a masked array (provided the mask isn’t a hard mask), the simplest solution is to
assign the constant nomask to the mask:
>>> x = ma.array([1, 2, 3], mask=[0, 0, 1])
>>> x
masked_array(data = [1 2 --],
mask = [False False True],
fill_value = 999999)
>>> x.mask = ma.nomask
>>> x
masked_array(data = [1 2 3],
mask = [False False False],
fill_value = 999999)

Indexing and slicing

As a MaskedArray is a subclass of numpy.ndarray, it inherits its mechanisms for indexing and slicing.
When accessing a single entry of a masked array with no named fields, the output is either a scalar (if the corresponding
entry of the mask is False) or the special value masked (if the corresponding entry of the mask is True):
>>> x = ma.array([1, 2, 3], mask=[0, 0, 1])
>>> x[0]
1
>>> x[-1]
masked_array(data = --,
mask = True,
fill_value = 1e+20)
>>> x[-1] is ma.masked
True

If the masked array has named fields, accessing a single entry returns a numpy.void object if none of the fields are
masked, or a 0d masked array with the same dtype as the initial array if at least one of the fields is masked.
>>> y = ma.masked_array([(1,2), (3, 4)],
... mask=[(0, 0), (0, 1)],
... dtype=[(’a’, int), (’b’, int)])
>>> y[0]
(1, 2)
>>> y[-1]
masked_array(data = (3, --),
mask = (False, True),
fill_value = (999999, 999999),
dtype = [(’a’, ’<i4’), (’b’, ’<i4’)])

270 Chapter 1. Array objects

NumPy Reference, Release 2.0.0.dev-ac3cba3

When accessing a slice, the output is a masked array whose data attribute is a view of the original data, and whose
mask is either nomask (if there was no invalid entries in the original array) or a copy of the corresponding slice of the
original mask. The copy is required to avoid propagation of any modification of the mask to the original.
>>> x = ma.array([1, 2, 3, 4, 5], mask=[0, 1, 0, 0, 1])
>>> mx = x[:3]
>>> mx
masked_array(data = [1 -- 3],
mask = [False True False],
fill_value = 999999)
>>> mx[1] = -1
>>> mx
masked_array(data = [1 -1 3],
mask = [False True False],
fill_value = 999999)
>>> x.mask
array([False, True, False, False, True], dtype=bool)
>>> x.data
array([ 1, -1, 3, 4, 5])

Accessing a field of a masked array with structured datatype returns a MaskedArray.

Operations on masked arrays

Arithmetic and comparison operations are supported by masked arrays. As much as possible, invalid entries of a
masked array are not processed, meaning that the corresponding data entries should be the same before and after the
operation.

Warning: We need to stress that this behavior may not be systematic, that masked data may be affected by the
operation in some cases and therefore users should not rely on this data remaining unchanged.

The numpy.ma module comes with a specific implementation of most ufuncs. Unary and binary functions that have a
validity domain (such as log or divide) return the masked constant whenever the input is masked or falls outside
the validity domain:
>>> ma.log([-1, 0, 1, 2])
masked_array(data = [-- -- 0.0 0.69314718056],
mask = [ True True False False],
fill_value = 1e+20)

Masked arrays also support standard numpy ufuncs. The output is then a masked array. The result of a unary ufunc is
masked wherever the input is masked. The result of a binary ufunc is masked wherever any of the input is masked. If
the ufunc also returns the optional context output (a 3-element tuple containing the name of the ufunc, its arguments
and its domain), the context is processed and entries of the output masked array are masked wherever the corresponding
input fall outside the validity domain:
>>> x = ma.array([-1, 1, 0, 2, 3], mask=[0, 0, 0, 0, 1])
>>> np.log(x)
masked_array(data = [-- -- 0.0 0.69314718056 --],
mask = [ True True False False True],
fill_value = 1e+20)

1.6. Masked arrays 271

NumPy Reference, Release 2.0.0.dev-ac3cba3

1.6.3 Examples

Data with a given value representing missing data

Let’s consider a list of elements, x, where values of -9999.