Support for numpy matrix operations

[alexfeld]

A few things regarding matrix operations that aren't working properly that would be useful.

• support for np.dot e.g. A.dot(b) or b.dot(A). Currently, if A,b are arrays containing quantities, this works. However, if A,b are Quantity's, then this operation does not return the correct result. I believe this is related to numpy: matrix multiplication by left give wrong results #39 .
• support for np.linalg.solve. Solving a system where A,b are instances of Quantity the method fails with a TypeError in line 1235 of quantity. I think this is related to Prevent crash when context is None #420. If A,b are arrays of Quantitys the method fails in linalg with a TypeError: no loop matching the specified signature was found for ufunc solve1

[amine-aboufirass]

Has this been implemented? The documentation states that Pint supports numpy array operations. A dot product is one of the most basic array operations. Otherwise I don't believe there is a point in supporting numpy....

We could implement a temporary solution like this until the more fundamental solutions are resolved:

def pint_dot(arr1, arr2):
    units_arr1 = arr1.units
    units_arr2 = arr2.units
    arr3 = np.dot(arr1.magnitude, arr2.magnitude)
    units_arr3 = arr1.units*arr2.units
    return arr3*units_arr3

a = np.array([1,2,3])*ureg.m
b = np.array([2,3,4])*ureg.m

pint_dot(a,b).to('cm**2')

>>> 200000.0 centimeter ** 2

[hgrecco]

Revisit after #905

[jthielen]

np.dot has been implemented in #905, but np.linalg.solve has not. I suspect it may be as simple as adding

for func_str in ['solve']:
    implement_func('function', func_str, input_units=None, output_unit='div')

to numpy_func.py, but it may be complicated by the fact that np.linalg.solve is in a submodule.