How to use TimeSeriesDataSet without using DataLoader

[jobs-git]

How to use TimeSeriesDataSet without using DataLoader? Of course I need the pre-processing being incorporated before the data is received by the model. For example in nbeats under _nbeats.py file, the forward function receives a dictionary:

def forward(self, x: Dict[str, torch.Tensor]) -> Dict[str, torch.Tensor]

in my understanding, that dictionary contains data that been processed from the original data frame. I think TimeSeriesDataSet does that? But I need those x data, skipping the dataloader batching. I am aming to perform manual batching. This is also related to: #1846

[fkiraly]

FYI @phoeenniixx, @PranavBhatP, @fnhirwa - I think you have to use the loader in the v1 design, but I am not 100% sure.