Hyper-param optimization giving suboptimal results compared to non-optimized models about neuralforecast HOT 2 CLOSED

Do I need to specify unique_ids somehow when calling the Auto_model?

No, as long as they are specified in the unique_id column (or whatever name you've given to that column) it should be fine.

If I do hyper-param optimization on the train_data using Auto_DilatedRNN having single unique_id, I get an error lower than the non-hyper-optimized model (i.e., the regular DilatedRNN). And this is expected (and the goal of hyper-param optimization).

However, when I call the Auto_model (i.e., AutoDilatedRNN) on the train dataset having two unique_id, I end up with a higher error than if I just called the non-optimized model (i.e. regular DilatedRNN) for one unique_id and the opposite for the other.

Can you please guide me about this? Many thanks!

That can happen. Not sure what the question / issue is? If you want a single neural network that can forecast multiple time series, use a single neural network on a dataset with multiple unique ids. If you want multiple neural networks (for each unique_id a single NN), you can do that too. There's no guarantee that one method will perform better than the other.

Personally I'd never use a separate NN for every time series, as it kind of defeats the purpose of using a NN. Is it possible that you may end up with a lower error by training separate NNs for every series? Sure. Does it make sense to do that? In most cases, no, not really.

Also, this seems like a toy setup - in reality you'd train a single NN on tens or hundreds or more timeseries, and that would commonly outperform the variant where you'd train a separate NN for every timeseries. But for just two timeseries, sure, it can go both ways.

from neuralforecast.

@elephaint wanted to follow up on this. Would highly appreciate any guidance. Many thanks!

from neuralforecast.