Using np.linalg.inv with unyt_array has no effect on units

[sharkweek]

• unyt version: 1.0.7
• Python version: 3.6
• Operating System: Windows 10

Description

Using the numpy.linalg.inv() to calculate the inverse of a matrix, doesn't affect units.

What I Did

Here is what happens when you take the inverse of an unyt_array:

>>> x = unyt.unyt_array([[1. , 0.5, 0.5],
                         [0.5, 1. , 0.5],
                         [0.5, 0.5, 1. ]],
                        'psi')
>>> np.linalg.inv(x)
unyt_array([[ 1.5, -0.5, -0.5],
            [-0.5,  1.5, -0.5],
            [-0.5, -0.5,  1.5]], 'psi')

numpy correctly carries out the inverting operation, but the units remain the same where they should now be the reciprocal of the original units ('inch**2/lbf' in this case). As of now, the only work around I can think of is to create a new unyt_array with the inverted value of the original one and the reciprocal of the original's units.

[ngoldbaum]

So I doubt there's anything we can do about making np.lingalg.inv respect units correctly without support in NumPy for __array_function__, which is being added in a future numpy release. I don't see any mention of lingalg.inv in NEP-18 but maybe @shoyer knows more?

We could also define a wrapper around lingalg.inv that correctly handles units and add the wrapper to unyt.

[shoyer]

Every function that accepts arrays or array-like objects in NumPy's public APIs will support NEP 18. That includes np.linalg.inv.

[sharkweek]

That's good news. I suppose we just have to wait until the numpy release that implements __array_function__.