Hi,
Have you ever tried to build dense stacked LSTM on Keras?
I am trying to reproduce the dense big two-stacked LSTM on PTB dataset on keras. With learning rate = 1, the perplexity can reach 220, however, neither using decay 0.1 as your code or the original decay 1/1.15 as the original code can not decrease the perplexity any more.
Could you please provide any tips for this problem? Thanks.
Hi, Have you ever tried to build dense stacked LSTM on Keras? I am trying to reproduce the dense big two-stacked LSTM on PTB dataset on keras. With learning rate = 1, the perplexity can reach 220, however, neither using decay 0.1 as your code or the original decay 1/1.15 as the original code can not decrease the perplexity any more. Could you please provide any tips for this problem? Thanks.