numpy: matrix multiplication by left give wrong results

[canesin]

Hi, this is a issue related to #37 , multiplication by left don't work with a numpy matrix, as matrices are not commutative for multiplication. It don't fail, it give wrong results.

https://gist.github.com/canesin/6115159

[hgrecco]

Thanks again for the report. I am looking for a good solution for this.

[canesin]

Any news on this ?? I discovered that maybe this affects other operations also, like comparisons.

[hgrecco]

I have a working solution for #37 and #39 that involves refactoring everything so that Quantity dervies from several numpy classes. Briefly:

class Quantity(np.matrixlib.defmatrix.matrix, np.ma.core.MaskedArray, np.ndarray):
# here goes the class definition.

I find this extremely ugly, but seems to be the only thing that works. The problem is related to the fact that matrix and masked array take preference in spite of the the array priority.

[canesin]

It has other implications besides being ugly ?? You could push the fix as a temporary and refactor in the future, maybe create a enhancement issue for that.

[akirkby]

I'm having the same issue doing numpy matrix operations while using Pint... any news on this fix?

[hgrecco]

I have just pushed something I was working on to a temporary branch named _npsubclass. I putting this in the wild to get some comments and ideas. The main change is that Quantity now is a subclass from ndarray (which is an alias for object when NumPy is not available). This solves the problem of Quantity not being recognized by as array but introduces other.

I would like to emphasize that this is very preliminary work. We still do not know if this is the right approach or even if it is a good approach. Therefore we cannot promise that I will become part of Pint.

[akirkby]

Thanks Hernan, I'll check it out

Sent from my iPhone

On Jul 8, 2014, at 10:50 AM, Hernan Grecco [email protected] wrote:

I have just pushed something I was working on to a temporary branch named _npsubclass. I putting this in the wild to get some comments and ideas. The main change is that Quantity now is a subclass from ndarray (which is an alias for object when NumPy is not available). This solves the problem of Quantity not being recognized by as array but introduces other.

I would like to emphasize that this is very preliminary work. We still do not know if this is the right approach or even if it is a good approach. Therefore we cannot promise that I will become part of Pint.

—
Reply to this email directly or view it on GitHub.

[hgrecco]

Review once that #905 is implemented

[jthielen]

Based on https://docs.scipy.org/doc/numpy-1.17.0/reference/generated/numpy.matrix.html, NumPy matrices should no longer be used, instead, regular arrays should be used. So, can this be closed as wontfix due to upstream deprecation?

[hgrecco]

Agreed!