yangysc / wavelet-hfcm Goto Github PK

During the training process, the steepness value of tanh function is 1 by default, however, in prediction process, the steepness values are set as :
steepness = np.max(np.abs(W_learned), axis=1)
for i in range(Nc):
if steepness[i] > 1:
W_learned[i, :] /= steepness[i]

I am confused about this. Because it is not consistent with training target. Looking forward to your reply.

hello , i got a question and confuse me a lot. can i get email address?
thanks ……^_^

def wavelet_transform(x, J):
    N = len(x)
    C = np.zeros(shape=(J + 1, N))
    # W: wavelet coefficients
    W = np.zeros(shape=(J + 1, N))
    C[0, :] = x.copy()
    for j in range(1, J + 1):
        for k in range(1, N):
            C[j, k] = 1 / 2 * (C[j - 1, k] + C[j - 1, k - np.power(2, j - 1)])
            W[j, k] = C[j - 1][k] - C[j, k]

    W[0, :] = C[J, :]
    return W[:, np.power(2, J):] 

你好，为什么系数在返回时只选取了2^J以后的，这一点让我感觉很疑惑，希望您有时间可以为我解答。十分感谢

Thanks for sharing the codes which are very helpful.

However, I fail to generate the same results as in the paper (RMSE), although I set the same hyper-parameters.