why my loss_shape is negative

[Firvingxc]

i have a question about my loss_shape, i don't know why the value is negative. i think it's not right, but i don't know why

[camel712]

I have the same question, did you solve it?

[MantautasRimkus]

I have noticed the same thing. Especially, if I provide matrix with all zeroes, loss shape still comebacks as negative. In function

@njit
def compute_softdtw_batch_channel(D, gamma):
    batch_size = D.shape[0]
    num_channels = D.shape[1]
    N = D.shape[2]
    M = D.shape[3]
    R = np.zeros((batch_size, num_channels, N + 2, M + 2), dtype=np.float32) + 1e8
    R[:, :, 0, 0] = 0
    for j in range(1, M + 1):
        for i in range(1, N + 1):
            r0 = -R[:, :, i - 1, j - 1] / gamma
            r1 = -R[:, :, i - 1, j] / gamma
            r2 = -R[:, :, i, j - 1] / gamma
            rmax = np.maximum(np.maximum(r0, r1), r2)
            rsum = np.exp(r0 - rmax) + np.exp(r1 - rmax) + np.exp(r2 - rmax)
            softmin = - gamma * (np.log(rsum) + rmax)
            R[:, :, i, j] = D[:, :, i - 1, j - 1] + softmin

    return R

I have noticed the addition of rmax to np.log(rsum) - I do not see such action in the paper and if |np.log(rsum)|<rmax, softmin becomes negative. I was wondering if this could be reason by loss shape is negative