Greetings
I’m trying to do classification by using LSTM network.
However, the output is easily biased - it means the output looks like
0.3 0.7
0.3 0.7
0.3 0.7
0.3 0.7
0.3 0.7
for the batch size with 5. (Actually I’m doing it with the size of 128)
To observe the dramatic change, I modified the learning rate from 1e-5 to 1e-1.
The result shows like this
0.3 0.7 -> 0.6 0.7
0.3 0.7 -> 0.6 0.7
0.3 0.7 -> 0.6 0.7
0.3 0.7 -> 0.6 0.7
0.3 0.7 -> 0.6 0.7
Shortly, the output for every batch data moves together and the training is totally failed.
Also when I try to print out the gradient of variables, None is printed out.
If you want more details, I can upload the whole code or github address.
Thank you.