Training two layer networks fails with gradient errors

Question

Training two layer networks fails with gradient errors

mihaelacr-google opened this issue 8 years ago · 2 comments

I have added a createBidi2 method in the CLSTMText class function, but training 2 layer networks fails with a gradient assertion error, regardless of the used learning rate and hidden layer sizes.

My code to create two layer here.

The error I am getting is
clstmfiltertrain: clstm.cc:642: void ocropus::GenericNPLSTM<F, G, H>::backward() [with int F = 1; int G = 2; int H = 2]: Assertion `!anynan(out)' failed.

Looks like there is an error when computing the gradient of two layer networks.

Answer 1 · 2017-04-19T09:48:21.000Z

I am facing the same problem.

Answer 2 · 2017-10-04T13:15:57.000Z

If this can be of any help to someone, I got the exact same error when I run the clstmocrtrain without the corresponding *.gt.txt files in the image folder.