tutorials1-FinRL_PortfolioAllocation_NeurIPS_2020 can't train, it outputs "nan"

[chenxiao1129]

i think i have the right data

but when at part6, i failed to train_model

i try to find the problem
it seems here where "features" is no "nan"
but after "mlp_extractor()", the "latent_pi" is a "nan" tensor

is there anybody know why this happen? and how to deal with it?

[Athe-kunal]

It seems like your model has diverged and while extracting policy using log, it leads to NaN values. You can try hyperparameters optimization, maybe the current ones are suboptimal and diverging the models. The tutorial for the same is here