search

knowledgetechnologyuhh / OMGEmotionChallenge

Repository for th OMG Emotion Challenge
Apache License 2.0
87 stars 36 forks source link

The score on training set is very different from the one on validation set #4

Closed ewrfcas closed 6 years ago

ewrfcas commented 6 years ago

I can get 0.78 score in training set while the one is only 0.04 on validation set. What's wrong with my method? Who can give me some advices?

EgorLakomkin commented 6 years ago

Hi @ewrfcas and thank you for your interest in this challenge! Maybe the gap you observe can be due to model overfitting and some regularization methods can help (various variants of dropout, batch/layer/weight normalization). Do you monitor the model performance during training on the held out dataset?

ewrfcas commented 6 years ago

@EgorLakomkin Thank you for your advices. Unfortunately, the loss of the held out dataset achieves no improvements after the first epoch.

pablovin commented 6 years ago

Dear @ewrfcas , we had some problems with the cccalculation script, could you update the script and try it again now? maybe you get better results.