Can DeepVar handle multiple multivariates series

[stailx]

• PyTorch-Forecasting version: 0.10.2
• Python version: 3.9.7

Hello, I am currently trying to use DeepVAR for a dataset containing multiple series that each contains several correlated targets, I thought I could use this implementation but I am struggling to make it work.

As far as I understood on the tutorial notebook for deepAR, the different series are considered as independent series (group_ids=["series"]) but because the batches are synchronized somehow the model can infer correlation between series. However, in my use case I do want to group by series and by target, but I only want to look for correlation between my targets and not between my series. As a result, I am wondering if there is a way to enforce this behaviour.

I was thinking of grouping only by targets and transform my different series to different targets in the dataset, to be able to use MultiLoss but it would constraint that the each series has the same length which is not the case.

Thank you very much in advance.