NLP model Problem

Course Q&A Deep Learning Specialization Sequence Models
1

Hello. I implement one NLP model with RNN, LSTM and GRU individually. My problem is that I get 99% train accuracy but 30% test accuracy. Can someone help me to understand what may can be my problem?

2

Sounds like you have overfit the training set.

1 Like
3

I think my problem is with the inputs of the LSTM, RNN and GRU models.

4

Yeah the inputs maybe not be suitable and not normalized to a distribution where a training set similar to test set is choosen. Other manipulations might also be need prior to inputing data to the NN and those same processing actions need to be implemented at the dev and test sets respectively. You should make sure you have one pipeline that all the data follow nomatter if is train, test or dev data.

1 Like