Units with matrix inversion

When inverting a matrix with units using numpy.linalg.inv, the units of the inverted matrix stay the same as the input matrix. The units should also be inverted. As an example:

import numpy
import astropy.units
scale = numpy.matrix([[-1, 0], [0, 1]])*astropy.units.arcsec/astropy.units.pixel
numpy.linalg.inv(scale)

The returned value should be in units of pix/arcsec but are instead kept as arcsec/pix.

[mhvk]

Thanks for reporting!

Indeed, np.matrix is not supported with units (it is a different np.ndarray subclass). Note that it is also deprecated on the numpy side, where it is suggested just to use arrays with the relevant shape.

In principle, I would hope that we can eventually support linalg with quantity arrays (with matrix shape), but we’re not there yet (needs implementation of the new __array_function__, and numpy >=1.17. So, I fear that for now you’ll just have to do things by hand…

[bsipocz]

@mhvk - I’m removing this from the milestone, but keeping the big tracking one in, just a little bit longer.