Refactor weight gradients calculation

[selfint]

How to: (roughly)

Inputs:

• current layer errors (1d array with length N)
• previous layer activations (1d array with length M)

Output:

• NxM matrix with weight gradients for each weight in the current layer

Calculation: errors * prev_activations.T

This yields an NxM matrix (Nx1 • 1xM = NxM)

[selfint]

works, but maybe there is a better way to use ndarray to perform the transposes (instead of using from_shape_vec)