relative entropy using scipy.stats.distribution.entropy when input arrays have ndim > 1

[argriffing]

This seems to be broken.

$ python
Python 2.7.5+ (default, Sep 19 2013, 13:48:49) 
[GCC 4.8.1] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy as np
>>> import scipy.stats.distributions
>>> a = np.random.rand(2, 3)
>>> scipy.stats.distributions.entropy(a)
array([ 0.60673183,  0.27477313,  0.56533111])
>>> scipy.stats.distributions.entropy(a, a)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/argriffi/.local/lib/python2.7/site-packages/scipy/stats/distributions.py", line 6130, in entropy
    if any(take(pk, nonzero(qk == 0.0), axis=0) != 0.0, 0):
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

[josef-pkt]

Yes, a bug given that the docstring indicates sums are take with axis=0

Also we should get inf only in those positions where the column satisfies the condition.

[josef-pkt]

I think it should be something like

inf_mask = ((qk == 0) & (pk > 0)).any(0)
and if S is not scalar
S[inf_mask] = np.inf

[WarrenWeckesser]

Fixed by #3041