Make a separate preprocessor for Tabular representations

[saleml]

For now, Tabular representations require the Identity preprocessor. The preprocessed states (which are raw tensors, given that they are the output of the Identity preprocessor) are passed back to the environment to create States objects, on which get_states_indices is called. This is highly inefficient, and the code could benefit from having a TabularPreprocessor that calls get_states_indices under the hood. The output of TabularPreprocessor would thus be indices that can readily be used for any lookup table.

[danjenson]

Without having dived too deep into this yet, should we be using Dataset/DataLoader/Sampler classes built into torch, or at least follow their API design, which might be a bit more battle tested?

[saleml]

I'm not sure I understand your question @danjenson. How does this relate to Datasets and Dataloaders of PyTorch ? Maybe I didn't describe the issue clearly.