for循环中numpy数组的计算结果错误

Question

I have the following code that computes a wrong number in python with numpy array claimed already:

for t in range(T-1,0,-1):
    dh[:,t,:] = dh[:,t,:] + dprev_ht
    dx[:,t,:], dprev_ht, dWxt, dWht, dbt = rnn_step_backward(dh[:,t,:], cache[t])
    dWx += dWxt
    dWh += dWht
    db += dbt

But if I change it to:

for t in range(T-1,0,-1):
    cur_dh = dh[:,t,:] + dprev_ht
    dx[:,t,:], dprev_ht, dWxt, dWht, dbt = rnn_step_backward(cur_dh, cache[t])
    dWx += dWxt
    dWh += dWht
    db += dbt

Then I can get the correct number. Is this problem related to the order or precedence of computation in python or the mutability of numpy array in for loop? Or am I totally wrong about this question?