The training procedure of probabilistic time series forecasting is wrong 🔴

[oublalkhalid]

Hello, I wanted to bring to your attention an issue I encountered while working with the notebook provided for training time series models. The results I obtained do not match those mentioned in the table, and I'm unsure why.

One aspect that particularly confused me is the validation procedure. It seems that only the last sequence, which has been seen during training, is used for validation. While I attempted to create my own dataloader, the results are still far from those presented in the table.

I would appreciate it if you could provide some clarification on the validation process. Have the results been validated in scenarios where the context is never seen during training?

Thanks, Koublal