median of np.matrix is broken

[argriffing]

>>> import numpy as np                                                          
>>> a = np.matrix([[0, 1, 2]])                                                  
>>> np.mean(a)                                                                  
1.0
>>> np.median(a)                                                                
Traceback (most recent call last):
[...]
/numpy/core/fromnumeric.py", line 619, in partition
    a.partition(kth, axis=axis, kind=kind, order=order)
ValueError: kth(=1) out of bounds (1)

[empeeu]

This actually seems to be a problem with partition.

>>> np.median(a, axis=0)
matrix([[ 0.,  1.,  2.]])
>>> np.median(a, axis=1)
matrix([[ 1.]])
>>> np.partition(a, 1, axis=1)
matrix([[0, 1, 2]])
>>> np.partition(a, 1, axis=None)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python2.7/dist-packages/numpy/core/fromnumeric.py", line 619, in partition
    a.partition(kth, axis=axis, kind=kind, order=order)
ValueError: kth(=1) out of bounds (1)
>>> np.partition(np.array(a), kth, axis=None)
array([0, 1, 2])

median just so happens to use partition as part of its calculation.

[charris]

@juliantaylor I'm going to remove this as a blocker, but a fix would be nice. We can backport it later.

[juliantaylor]

I don't really know how to fix it, should matrix not override flatten to not flatten, or should squeeze be overriden to really flatten?

[charris]

Heh, still present in 1.12.

[dipsivenkatesh]

Reproducing Bug:

This bug seems to exist when we compute median for NumPy matrix, and that too when we don't specify the axis.

a = np.matrix([[0,1,2],[3,4,5]])
>>> np.median(a)
matrix([[0., 1., 2., 3., 4., 5.]])
>>> np.median(a,axis = 1)
matrix([[1.],
        [4.]])
>>> np.median(a,axis = 0)
matrix([[1.5, 2.5, 3.5]])

for numpy array, it flattens the array and computes the median but for the matrix it's not.