Giters

etzinis / sudo_rm_rf

Code for SuDoRm-Rf networks for efficient audio source separation. SuDoRm-Rf stands for SUccessive DOwnsampling and Resampling of Multi-Resolution Features which enables a more efficient way of separating sources from mixtures.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

loss is nan

weimisa341 opened this issue · comments

commented

when I use another dataset to train you model, I found that loss will become nan. I can use my dataloader to train other model in normal state. If I use mse loss to train your model, it can be trained. Could you please tell me your idea about this problem? Could you please tell me your normalized range？ I think it maybe a factor.

commented

Hey, as you realise I cannot possibly find the bug without looking at your code and your data. But I would probably guess that there is an all zeros vector somewhere that causes the log to overflow

commented

Hey, as you realise I cannot possibly find the bug without looking at your code and your data. But I would probably guess that there is an all zeros vector somewhere that causes the log to overflow

Thanks for your help