JamesTuna

Noise estimation in signal_matrix.py

13

Line: self.noise = np.std(diff) * 0.5 seems to be inconsistent with Frobenius term used in paper: ||M1-M2||/(2sqrt(mn)). Instead it should be np.sum(diff**2)**(0.5)/(2*n)? Btw, isn't m=n=vocabulary size? Under this assumption, what's...