search

piEsposito / blitz-bayesian-deep-learning

A simple and extensible library to create Bayesian Neural Network layers on PyTorch.
GNU General Public License v3.0
918 stars 107 forks source link

Loss issue #115

Open lweidsab opened 1 year ago

lweidsab commented 1 year ago

Hi,

When I use the Bayesian LSTM to construct a network as the ipython given, I use the customed temperature data and found the loss cannot decrease during the training process. image The original data is preprocessed by scaler function like this: image Does anyone come across this issue, I would be so so grateful if you could help.

Best regards, Yong

ynashed commented 1 year ago

@lweidsab Did you try to tune the KL divergence weight complexity_cost_weight? It maybe set too high.