I've noticed that the RuntimeWarning: divide by zero encountered in log warnings gets triggered for masked arrays as well.

Consider the following two examples:

Regular arrays

>>> import numpy as np
>>> y = np.zeros((3,3))   np.eye(3,3)
>>> np.log(y)
<stdin>:1: RuntimeWarning: divide by zero encountered in log
array([[  0., -inf, -inf],
       [-inf,   0., -inf],
       [-inf, -inf,   0.]])

Masked arrays

>>> import numpy as np
>>> y = np.zeros((3,3))   np.eye(3,3)
>>> y_mask = np.ma.masked_equal(y, 0)
>>> np.log(y_mask)
<stdin>:1: RuntimeWarning: divide by zero encountered in log
masked_array(
  data=[[0.0, --, --],
        [--, 0.0, --],
        [--, --, 0.0]],
  mask=[[False,  True,  True],
        [ True, False,  True],
        [ True,  True, False]],
  fill_value=0.0)

Is this the expected behaviour? I would have expected the log operation to only happen on the non zero values in the case of masked arrays.

Additionally I have noticed that the this warning only gets triggered once per session. Any subsequent calls to the log function whether on masked or regular arrays won't trigger this warning.

I could suppress this warning via numpy.seterr(divide = 'ignore') but I'm curious as to why the operation on masked arrays raises a warning in the first place.

CodePudding user response：

np.ma versions of numpy functions can do one of several things.

1. use compress to operate on just the unmasked values,

2. use fill to replace masked elements with something like 1 or 0,

3. use errstate to surpress the warnings. np.ma.log uses the surpress method

Source for np.ma.log is

Signature:       np.ma.log(a, *args, **kwargs)
Type:            _MaskedUnaryOperation
String form:     Masked version of <ufunc 'log'>
File:            /usr/local/lib/python3.8/dist-packages/numpy/ma/core.py
Source:         
class _MaskedUnaryOperation(_MaskedUFunc):
    """
    Defines masked version of unary operations, where invalid values are
    pre-masked.
    ...
    def __call__(self, a, *args, **kwargs):
    ...
         with np.errstate(divide='ignore', invalid='ignore'):
                result = self.f(d, *args, **kwargs)

Previously

Operations on numpy masked array gives invalid values masked