l1 regularization

[forcoogle]

Shouldn't class l1_regularization() be defined as self.alpha * np.linalg.norm(w, 1) instead of self.alpha * np.linalg.norm(w)?

[cpy18727]

i think so too

[klintan]

Yes, I was looking at that for way too long. It should def be np.linalg.norm(w, 1) for the L1-norm or Manhattan distance, the default is the L2-norm (Frobenius).