Odd behavior of ma.add.reduce on boolean arrays

[abalkin]

Consider two masked arrays:

In [40]: a = np.ma.array([True] * 3, mask=[False] * 3)

In [41]: b = np.ma.array([True] * 3)

The only difference is that one has an all-false mask and the other has mask=nomask. However, ma.add.reduce returns different values:

In [42]: np.ma.add.reduce(a)
Out[42]: True

In [43]: np.ma.add.reduce(b)
Out[43]: 3

[charris]

Hmm, looks like this might be a problem in ndarray also

In [6]: a = np.array([True] * 3)

In [7]: a + a
Out[7]: array([ True,  True,  True], dtype=bool)

In [8]: add.reduce(a)
Out[8]: 3

In [9]: add.reduce(a, dtype=bool_)
Out[9]: True

[abalkin]

@charris - the ndarray behavior that you show seems to follow naturally from the definition of addition on booleans: basically "add" means "or". Not the best, but it is what it is. What do you see as a bug in your display?

[abalkin]

The issue that I am reporting here is ma specific. Take a look at this if statement in the definition of reduce(). The dtype is handled differently depending on whether the mask is nomask or not.

[charris]

@abalkin The reference was to the fact that addition results in a boolean, but reduction returns and integer. That is an inconsistancy. I suspect both should return integer for addition, as a lot of code adds up the booleans for counting purposes, while, if the logical form is needed, logical_or is the way to go.

[abalkin]

@charris - this is documented for np.sum:

In [5]: np.sum?
Signature: np.sum(a, axis=None, dtype=None, out=None, keepdims=False)
..
dtype : dtype, optional
    The type of the returned array and of the accumulator in which
    the elements are summed.  By default, the dtype of `a` is used.
    An exception is when `a` has an integer type with less precision
    than the default platform integer.  In that case, the default
    platform integer is used instead.

Maybe we should document an exception for the reduce as well.

[lrq3000]

Related: https://www.noveltech.dev/booleans-python-numpy/