(X, Z, T, Y, W2)
| 46 | |
| 47 | |
| 48 | def derivative_w1(X, Z, T, Y, W2): |
| 49 | # dZ = np.outer(T-Y, W2) * Z * (1 - Z) # this is for sigmoid activation |
| 50 | # dZ = np.outer(T-Y, W2) * (1 - Z * Z) # this is for tanh activation |
| 51 | dZ = np.outer(T-Y, W2) * (Z > 0) # this is for relu activation |
| 52 | return X.T.dot(dZ) |
| 53 | |
| 54 | |
| 55 | def derivative_b1(Z, T, Y, W2): |
no outgoing calls