Overfitting Issue
anuraggoel70 opened this issue · 1 comments
anuraggoel70 commented
I tried to use the LSTM architecture in a emotion classification dataset having 13 different categories of emotion. But, the model is getting overfitted. I have tried many things to prevent it but no results yet. Training accuracy reaching upto 80% while Validation accuracy moving between 30-34% only. Any suggestions?
bentrevett commented
There's no "one size fits all" solution to overfitting.
How many examples do you have? Are your classes balanced? Have you tried adding more regularisation? Are you using pretrained embeddings like GloVe or a pretrained BERT provided by the transformers library?