Question

If you try the following code segment

import numpy as np
import numpy.ma as ma

a = np.random.random(100) + 1j*np.random.random(100)
mask = np.ones_like(a, dtype='bool')
mask[0:9] = False
a = ma.masked_array(a, mask)
phase = np.angle(a)

The phase array will not be masked. The angle function will return values for the whole array, even for the masked out values. Am I doing something wrong here or is this the way it should be? If so, why?

Answer 1

Had a quick look at the numpy source, and it might be a bug/not implemented yet.

It’s listed as a “missing feature (work in progress)” on the numpy.ma page, issue #1: http://projects.scipy.org/numpy/wiki/MaskedArray.

The problem is that a number of unary functions such as np.angle, np.quantile call [np.]asarray in the source, which strips out the mask.

As the devs explain in the page I linked to, if these functions used ma.asarray instead of np.asarray they’d work, but they don’t :(.

I guess this is a patch yet to be submitted?

As a temporary workaround, np.angle basically calls np.arctan2(a.imag,a.real) (optionally multiplying by 180/pi to get degrees), so you could use that.