Should “dc” not also be an input parameter for the “lstm_backward” function, as this is required for the “lstm_cell_backward” fuction call inside “lstm_backward”
Which variable do you put for dc_next parameter in the “lstm_cell_backward(da_next, dc_next, cache)”-call inside “lstm_backward”
Hi MH_Jazz, i have the same question. I saw someone use the initialized dc_prevt = np.zeros((n_a, m)) as dc_next to calculate the gradients. And it seems the result is right if you simply do so. But I cann’t figure out why. Can any mentor help here? Thanks a lot!
I had the same question. I used0’s and it worked, but I don’t understand why the partial derivative of loss with respect to memory cell ‘c’ should be 0’s if the partial derivative of loss with respect to short term memory ‘a’ is not…
Same question here. I know using dc_prevt is correct but still don’t know why. Why parameters contain da but not dc?
This thread has been cold for 11 months.
Please start a new thread for your question.