optimization

[jasmainak]

we should use vectorized operations., e.g., np.einsum or np.dot instead of np.sum(x * y)