If masked arrays are compared (through the >, >=, < and <= operators to constants, arrays or other masked arrays they access the masked out elements for comparison. In some extreme cases this fails, often causing a warning.

Reproducing code example:

This example

import numpy as np
d = np.ma.array([1, None], mask=[False, True])
d > 0

fails with a TypeError: '>' not supported between instances of 'NoneType' and 'int'.

This example

import numpy as np
d = np.ma.array([1, np.nan], mask=[False, True])
d > 0

throws a warning __main__:1: RuntimeWarning: invalid value encountered in greater.

I think masked out elements should be set to False by default and not compared, so that operations such as:

d[d > 0] = 0

have the expected behavior. However that might cause issues in other cases such as:

np.all(d > 0)

This issue has been brought up before here: #4959, but the initial issue there seems more about using default functions such as np.log with masked array insted of the np.ma relative.

Even adding a fill_value seems not to fix this. At least in that case I would assume that for elements that are masked out the comparison happens with respect to the fill value.

import numpy as np
d = np.ma.array([1, None], mask=[False, True], fill_value=-1)
d > 0

creates the same error as mentioned above.

Numpy/Python version information:

• Numpy: '1.18.2'
• System: '3.7.6 (default, Jan 30 2020, 09:44:41) [GCC 9.2.1 20190827 (Red Hat 9.2.1-1)]'