Skip to content

Commit 4b78dd6

Browse files
committed
Ensure bottleneck result types are consistent with numpy
1 parent a723956 commit 4b78dd6

1 file changed

Lines changed: 52 additions & 81 deletions

File tree

‎astropy/stats/sigma_clipping.py‎

Lines changed: 52 additions & 81 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,6 @@
11
# Licensed under a 3-clause BSD style license - see LICENSE.rst
22

3+
import functools
34
import warnings
45

56
import numpy as np
@@ -12,9 +13,6 @@
1213
from astropy.utils.compat.optional_deps import HAS_BOTTLENECK
1314
from astropy.utils.exceptions import AstropyUserWarning
1415

15-
if HAS_BOTTLENECK:
16-
import bottleneck
17-
1816
if NUMPY_LT_2_0:
1917
from numpy.core.multiarray import normalize_axis_index
2018
else:
@@ -23,66 +21,61 @@
2321
__all__ = ["SigmaClip", "sigma_clip", "sigma_clipped_stats"]
2422

2523

26-
def _move_tuple_axes_first(array, axis):
27-
"""
28-
Bottleneck can only take integer axis, not tuple, so this function
29-
takes all the axes to be operated on and combines them into the
30-
first dimension of the array so that we can then use axis=0.
31-
"""
32-
# Figure out how many axes we are operating over
33-
naxis = len(axis)
34-
35-
# Add remaining axes to the axis tuple
36-
axis += tuple(i for i in range(array.ndim) if i not in axis)
37-
38-
# The new position of each axis is just in order
39-
destination = tuple(range(array.ndim))
40-
41-
# Reorder the array so that the axes being operated on are at the
42-
# beginning
43-
array_new = np.moveaxis(array, axis, destination)
44-
45-
# Collapse the dimensions being operated on into a single dimension
46-
# so that we can then use axis=0 with the bottleneck functions
47-
array_new = array_new.reshape((-1,) + array_new.shape[naxis:])
24+
if HAS_BOTTLENECK:
25+
import bottleneck
4826

49-
return array_new
27+
def _move_tuple_axes_first(array, axis):
28+
"""
29+
Bottleneck can only take integer axis, not tuple, so this function
30+
takes all the axes to be operated on and combines them into the
31+
first dimension of the array so that we can then use axis=0.
32+
"""
33+
# Figure out how many axes we are operating over
34+
naxis = len(axis)
5035

36+
# Add remaining axes to the axis tuple
37+
axis += tuple(i for i in range(array.ndim) if i not in axis)
5138

52-
def _nanmean(array, axis=None):
53-
"""Bottleneck nanmean function that handle tuple axis."""
54-
if isinstance(axis, tuple):
55-
array = _move_tuple_axes_first(array, axis=axis)
56-
axis = 0
39+
# The new position of each axis is just in order
40+
destination = tuple(range(array.ndim))
5741

58-
if isinstance(array, Quantity):
59-
return array.__array_wrap__(bottleneck.nanmean(array, axis=axis))
60-
else:
61-
return bottleneck.nanmean(array, axis=axis)
42+
# Reorder the array so that the axes being operated on are at the
43+
# beginning
44+
array_new = np.moveaxis(array, axis, destination)
6245

46+
# Collapse the dimensions being operated on into a single dimension
47+
# so that we can then use axis=0 with the bottleneck functions
48+
array_new = array_new.reshape((-1,) + array_new.shape[naxis:])
6349

64-
def _nanmedian(array, axis=None):
65-
"""Bottleneck nanmedian function that handle tuple axis."""
66-
if isinstance(axis, tuple):
67-
array = _move_tuple_axes_first(array, axis=axis)
68-
axis = 0
50+
return array_new
6951

70-
if isinstance(array, Quantity):
71-
return array.__array_wrap__(bottleneck.nanmedian(array, axis=axis))
72-
else:
73-
return bottleneck.nanmedian(array, axis=axis)
52+
def _apply_bottleneck(function, array, axis=None, **kwargs):
53+
"""Wrap bottleneck function to handle tuple axis.
7454
55+
Also takes care to ensure the output is of the expected type,
56+
i.e., a quantity, numpy array, or numpy scalar.
57+
"""
58+
if isinstance(axis, tuple):
59+
array = _move_tuple_axes_first(array, axis=axis)
60+
axis = 0
61+
62+
result = function(array, axis=axis, **kwargs)
63+
if isinstance(array, Quantity):
64+
return array.__array_wrap__(result)
65+
elif isinstance(result, float):
66+
# For compatibility with numpy, always return a numpy scalar.
67+
return np.float64(result)
68+
else:
69+
return result
7570

76-
def _nanstd(array, axis=None, ddof=0):
77-
"""Bottleneck nanstd function that handle tuple axis."""
78-
if isinstance(axis, tuple):
79-
array = _move_tuple_axes_first(array, axis=axis)
80-
axis = 0
71+
_nanmean = functools.partial(_apply_bottleneck, bottleneck.nanmean)
72+
_nanmedian = functools.partial(_apply_bottleneck, bottleneck.nanmedian)
73+
_nanstd = functools.partial(_apply_bottleneck, bottleneck.nanstd)
8174

82-
if isinstance(array, Quantity):
83-
return array.__array_wrap__(bottleneck.nanstd(array, axis=axis, ddof=ddof))
84-
else:
85-
return bottleneck.nanstd(array, axis=axis, ddof=ddof)
75+
else:
76+
_nanmean = np.nanmean
77+
_nanmedian = np.nanmedian
78+
_nanstd = np.nanstd
8679

8780

8881
def _nanmadstd(array, axis=None):
@@ -279,16 +272,10 @@ def __str__(self):
279272
def _parse_cenfunc(cenfunc):
280273
if isinstance(cenfunc, str):
281274
if cenfunc == "median":
282-
if HAS_BOTTLENECK:
283-
cenfunc = _nanmedian
284-
else:
285-
cenfunc = np.nanmedian # pragma: no cover
275+
cenfunc = _nanmedian
286276

287277
elif cenfunc == "mean":
288-
if HAS_BOTTLENECK:
289-
cenfunc = _nanmean
290-
else:
291-
cenfunc = np.nanmean # pragma: no cover
278+
cenfunc = _nanmean
292279

293280
else:
294281
raise ValueError(f"{cenfunc} is an invalid cenfunc.")
@@ -299,10 +286,7 @@ def _parse_cenfunc(cenfunc):
299286
def _parse_stdfunc(stdfunc):
300287
if isinstance(stdfunc, str):
301288
if stdfunc == "std":
302-
if HAS_BOTTLENECK:
303-
stdfunc = _nanstd
304-
else:
305-
stdfunc = np.nanstd # pragma: no cover
289+
stdfunc = _nanstd
306290
elif stdfunc == "mad_std":
307291
stdfunc = _nanmadstd
308292
else:
@@ -316,15 +300,7 @@ def _compute_bounds(self, data, axis=None):
316300
with warnings.catch_warnings():
317301
warnings.simplefilter("ignore", category=RuntimeWarning)
318302
cen = self._cenfunc_parsed(data, axis=axis)
319-
if np.isscalar(cen) and not isinstance(cen, np.float64):
320-
# change bottleneck float scalar to np.float64
321-
cen = np.float64(cen)
322-
323303
std = self._stdfunc_parsed(data, axis=axis)
324-
if np.isscalar(std) and not isinstance(std, np.float64):
325-
# change bottleneck float scalar to np.float64
326-
std = np.float64(std)
327-
328304
self._min_value = cen - (std * self.sigma_lower)
329305
self._max_value = cen + (std * self.sigma_upper)
330306

@@ -1030,13 +1006,8 @@ def sigma_clipped_stats(
10301006
data, axis=axis, masked=False, return_bounds=False, copy=True
10311007
)
10321008

1033-
if HAS_BOTTLENECK:
1034-
mean = _nanmean(data_clipped, axis=axis)
1035-
median = _nanmedian(data_clipped, axis=axis)
1036-
std = _nanstd(data_clipped, ddof=std_ddof, axis=axis)
1037-
else: # pragma: no cover
1038-
mean = np.nanmean(data_clipped, axis=axis)
1039-
median = np.nanmedian(data_clipped, axis=axis)
1040-
std = np.nanstd(data_clipped, ddof=std_ddof, axis=axis)
1009+
mean = _nanmean(data_clipped, axis=axis)
1010+
median = _nanmedian(data_clipped, axis=axis)
1011+
std = _nanstd(data_clipped, ddof=std_ddof, axis=axis)
10411012

10421013
return mean, median, std

0 commit comments

Comments
 (0)