mstats.ttest_rel axis=None, requires masked array

[josef-pkt]

mstats.ttest_rel has wrong axis default axis=None instead of axis=0
requires masked array if axis=None

ma.ravel does not convert to masked arrays, at least not with numpy 1.6.1

def _chk2_asarray(a, b, axis):
    if axis is None:
        a = ma.ravel(a)
        b = ma.ravel(b)

>>> outcome = np.random.randn(20, 4) + [0, 0, 1, 2]
>>> stats.mstats.ttest_rel(outcome[:, 0], outcome[:, 1:])
Traceback (most recent call last):
  File "<pyshell#30>", line 1, in <module>
    stats.mstats.ttest_rel(outcome[:, 0], outcome[:, 1:])
  File "C:\Programs\Python27\lib\site-packages\scipy\stats\mstats_basic.py", line 782, in ttest_rel
    raise ValueError('unequal length arrays')
ValueError: unequal length arrays

>>> [stats.mstats.ttest_rel(outcome[:, 0], outcome[:, k]) for k in range(1, 4)]
Traceback (most recent call last):
  File "<pyshell#31>", line 1, in <module>
    [stats.mstats.ttest_rel(outcome[:, 0], outcome[:, k]) for k in range(1, 4)]
  File "C:\Programs\Python27\lib\site-packages\scipy\stats\mstats_basic.py", line 785, in ttest_rel
    n = a.count(axis)
AttributeError: 'numpy.ndarray' object has no attribute 'count'

[rgommers]

The issue with _chk2_asarray I had already fixed in gh-3008.

[rgommers]

Other issue now also fixed in gh-3008.