NumPy's pad with reflect_type="odd" appears strange in corners #11283
Description
Was trying to understand what NumPy's pad does with reflect_type="odd". According to the docs...
For the ‘odd’ style, the extented part of the array is created by subtracting the reflected values from two times the edge value.
So naively would think this is a composition of pad with mode="edge" and pad with mode="reflect"/mode="symmetric". Went ahead and tried this to see if this would work. Appears it almost does with the exception of corners (though edges appear correct).
While there are some comments in the docs about how corners are dealt with, it's still not obvious to me what is happening here unlike some of the other modes' corner behavior. Not sure if this is just me not understanding or if there is a genuine bug here. Please let me know your thoughts. Example of the behavior below.
Example:
In [1]: import numpy as np
In [2]: np.set_printoptions(linewidth=157)
In [3]: np.random.seed(0)
In [4]: a = np.random.random((3, 4))
In [5]: 2 * np.pad(a, 1, "edge") - np.pad(a, 1, "reflect", reflect_type="even")
Out[5]:
array([[0.45173289, 0.67397221, 0.78448462, 0.76793954, 0.19799337, 0.65217915],
[0.38243764, 0.5488135 , 0.71518937, 0.60276338, 0.54488318, 0.48700299],
[0.20141549, 0.4236548 , 0.64589411, 0.43758721, 0.891773 , 1.34595879],
[1.543884 , 0.96366276, 0.38344152, 0.79172504, 0.52889492, 0.2660648 ],
[1.28143141, 1.50367072, 0.12098892, 1.14586286, 0.16601684, 0.62020263]])
In [6]: np.pad(a, 1, "reflect", reflect_type="odd")
Out[6]:
array([[ 0.5634598 , 0.67397221, 0.78448462, 0.76793954, 0.19799337, -0.37195281],
[ 0.38243764, 0.5488135 , 0.71518937, 0.60276338, 0.54488318, 0.48700299],
[ 0.20141549, 0.4236548 , 0.64589411, 0.43758721, 0.891773 , 1.34595879],
[ 1.543884 , 0.96366276, 0.38344152, 0.79172504, 0.52889492, 0.2660648 ],
[ 2.88635252, 1.50367072, 0.12098892, 1.14586286, 0.16601684, -0.81382919]])
In [7]: (2 * np.pad(a, 1, "edge") - np.pad(a, 1, "reflect", reflect_type="even")) - (np.pad(a, 1, "reflect", reflect_type="odd"))
Out[7]:
array([[-0.1117269 , 0. , 0. , 0. , 0. , 1.02413197],
[ 0. , 0. , 0. , 0. , 0. , 0. ],
[ 0. , 0. , 0. , 0. , 0. , 0. ],
[ 0. , 0. , 0. , 0. , 0. , 0. ],
[-1.60492111, 0. , 0. , 0. , 0. , 1.43403182]])Environment:
name: test
channels:
- conda-forge
- defaults
dependencies:
- appnope=0.1.0=py36_0
- backcall=0.1.0=py_0
- blas=1.1=openblas
- ca-certificates=2018.4.16=0
- certifi=2018.4.16=py36_0
- decorator=4.3.0=py_0
- ipython=6.4.0=py36_0
- ipython_genutils=0.2.0=py36_0
- jedi=0.12.0=py36_0
- libgfortran=3.0.0=0
- ncurses=5.9=10
- numpy=1.14.4=py36_blas_openblash24bf2e0_200
- openblas=0.2.20=8
- openssl=1.0.2o=0
- parso=0.2.1=py_0
- pexpect=4.6.0=py36_0
- pickleshare=0.7.4=py36_0
- prompt_toolkit=1.0.15=py36_0
- ptyprocess=0.5.2=py36_0
- pygments=2.2.0=py36_0
- python=3.6.5=1
- readline=7.0=0
- setuptools=39.2.0=py36_0
- simplegeneric=0.8.1=py36_0
- six=1.11.0=py36_1
- sqlite=3.20.1=2
- tk=8.6.7=0
- traitlets=4.3.2=py36_0
- wcwidth=0.1.7=py36_0
- xz=5.2.3=0
- zlib=1.2.11=h470a237_3