Masked arrays don't properly edit views

I tried to change the mask of a through a subindexed view, but was unable. Using this setup I can reproduce this in the 1.9.1 version of NumPy.

import numpy as np

a = np.arange(6).reshape(2,3)
a = np.ma.masked_array(a, mask=np.ma.getmaskarray(a), shrink=False)

b = a[1:2,1:2]

c = np.zeros(b.shape, b.dtype)
c = np.ma.masked_array(c, mask=np.ma.getmaskarray(c), shrink=False)
c[:] = np.ma.masked

This yields what one would expect for a, b, and c (seen below).

 masked_array(data =
   [[0 1 2]
    [3 4 5]],
              mask =
   [[False False False]
    [False False False]],
         fill_value = 999999)

 masked_array(data =
   [[4]],
              mask =
   [[False]],
         fill_value = 999999)

 masked_array(data =
   [[--]],
              mask =
   [[ True]],
         fill_value = 999999)

Now, it would seem reasonable that to copy data into b from c one can use __setitem__ (seen below).

 b[:] = c

This results in new data and mask for b.

 masked_array(data =
   [[--]],
              mask =
   [[ True]],
         fill_value = 999999)

This should, in turn, change a. However, the mask of a remains unchanged (seen below).

 masked_array(data =
   [[0 1 2]
    [3 0 5]],
              mask =
   [[False False False]
    [False False False]],
         fill_value = 999999)

lazyflow/stype.py: Added a workaround for copying NumPy masked arrays…

lazyflow/slot.py: Added a workaround for copying NumPy masked arrays …

lazyflow/operators/opArrayCache.py: Workaround issues using masked ar…

lazyflow/stype.py: Added a workaround for copying NumPy masked arrays…

lazyflow/slot.py: Added a workaround for copying NumPy masked arrays …

lazyflow/operators/opArrayCache.py: Workaround issues using masked ar…

@jakirkham

@jakirkham, In your example code, I don't understand why you expect a.mask to be changed. The masks for all three arrays a, b, and c are False everywhere. Is the example missing a line such as c.mask[:] = True?

FWIW, If I insert that line, then I do indeed see the issue reported here. If I assign to a view of a masked array, the data values are updated in both the view and the original array (of course). But the mask is only properly updated in the view, not the original array.

System details:
numpy 1.9.1
python 2.7.9 (installed via Anaconda)
Mac OS 10.7.5

@stuarteberg

@stuarteberg Yes, sorry. You are correct I was missing such a line.

Generally, the documentation recommends one to assign as so c[:] = np.ma.masked. However, as we have ensured the mask is simply an array, it is also mentioned that you can do c.mask[:] = True as you have.

This code seems to stop propagation of the mask and it is not clear to me why this is necessary (

numpy/numpy/ma/core.py

Lines 3129 to 3132 in 4bfcf48

    
           # Unshare the mask if necessary to avoid propagation 
        
           if not self._isfield: 
        
               self.unshare_mask() 
        
               _mask = ndarray.__getattribute__(self, '_mask')

). As the mask is not hard here, it should not be a problem to have propagation.

Same situation here (

numpy/numpy/ma/core.py

Lines 3112 to 3113 in 4bfcf48

    
           if not self._isfield: 
        
               self._sharedmask = False

). I don't understand why only fields of masked record arrays can share masks, but nothing else. Is there something special we want to avoid in record arrays? If so, couldn't we accomplish this without ruining the possibility of editing views on an ordinary masked array?