Closed
Description
Something has changed between numpy 1.11 and the master branch, and it seems np.ma.median cannot handle an unmasked (mask=False) MaskedArray.
For example: np.ma.median(np.ma.array(np.ones((5, 5))), axis=0) throws an IndexError.
I haven't investigated it but to me it seems that np.ma.array creates a new mask (if the array hasn't got a mask already) and it is False (scalar). When using an axis with np.ma.median I suspect that the mask is applied to the data and then throws this Error:
IndexError Traceback (most recent call last)
<ipython-input-26-b5117c1f3918> in <module>()
----> 1 np.ma.median(np.ma.array(np.ones((5, 5))), axis=0)
C:\-\lib\site-packages\numpy\ma\extras.py in median(a, axis, out, overwrite_input, keepdims)
693
694 r, k = _ureduce(a, func=_median, axis=axis, out=out,
--> 695 overwrite_input=overwrite_input)
696 if keepdims:
697 return r.reshape(k)
C:\-\lib\site-packages\numpy\lib\function_base.py in _ureduce(a, func, **kwargs)
3631 keepdim = [1] * a.ndim
3632
-> 3633 r = func(a, **kwargs)
3634 return r, keepdim
3635
C:\-\lib\site-packages\numpy\ma\extras.py in _median(a, axis, out, overwrite_input)
736 low = high
737 else:
--> 738 low[odd] = high[odd]
739
740 if np.issubdtype(asorted.dtype, np.inexact):
C:\-\lib\site-packages\numpy\ma\core.py in __getitem__(self, indx)
3161
3162 """
-> 3163 dout = self.data[indx]
3164 # We could directly use ndarray.__getitem__ on self.
3165 # But then we would have to modify __array_finalize__ to prevent the
IndexError: in the future, 0-d boolean arrays will be interpreted as a valid boolean index
Which currently breaks some tests for astropy (see astropy/astropy#5303).