How does the splitter manage to retain the information contained in item_id, in the case of multidimensional series?

[c1adrien]

Hey !
In the following code, I am constructing a PandasDataset object containing a multidimensional series:

self.dataset = PandasDataset.from_long_dataframe(df,
                            item_id="item_id",
                             target="target",
                             feat_dynamic_real=["dynamic_item1","dynamic_item2","dynamic_item3"]) 

Then I apply a Transformation before executing the InstanceSplitter present in DeepVAR. However, I feel like InstanceSplitter does not retain the item_id information. This information is also not retained in the dataloader. So, if I pass a multidimensional series, how does the model distinguish between two series from two different item_ids during training? Thank you very much.