Numpy functions may not do what you think

Posted by Simon Walker on Fri 06 May 2016

Numpy has the ability to mask arrays and ignore their values for certain computations, called "masked arrays". They contain a .mask attribute which is a boolean array, True where the value should be masked and False otherwise.

Numpy also comes with a suite of functions which can handle this masking naturally. Typically for a function in the np. namespace, there is a masked-array-aware version under the np.ma. namespace:

np.median  => np.ma.median
np.average => np.ma.average

A crucial thing to remember however is that standard numpy functions ignore the mask for a masked array.

This caught me out when sigma clipping values using astropy.stats.sigma_clip - which masks out values outside the sigma range. To ignore the sigma clipped values I should have used np.ma.median instead of np.median.

tags: python, numpy