maciejkula / spotlight

Deep recommender models using PyTorch.
MIT License
3k stars 428 forks source link

Loss improves super slowly after first epoch in LSTM based sequential model. #168

Open snemistry opened 5 years ago

snemistry commented 5 years ago

LSTM based sequential model seems to fit data really well on first epoch, but validation accuracy doesn't improve after first epoch and even loss halts improving after second epoch. I tried different optimizer paramters like lr, l2, etc., but it doesn't improve beyond certain point.

@maciejkula Any idea how can improve that?