Add Tabnet support

[tztsai]

Description

I have added the TabNet model now. It is added as a building block in other models, similar to MLP. To incorporate the new model in current plugins like tabular_gan, tabddpm, etc., My approach is to remove the hyper-parameters specific to the MLP model and replace them with two parameters model_type and model_params. I have created a file "factory.py" that contains functions that given a name string and a dict of params, will return an nn module, or an activation layer, or a feature encoder. The get_model function, provided with model_type and model_params, can instantiate an NN block such as an MLP, TransformerModel, or TabNet. Consequently, by substituting MLP blocks with the dynamically instantiated blocks, we can provide more flexibility to the present plugins. I have tried using TabNet as the diffusion model of TabDDPM, instead of MLP, and it can successfully fit and generate data, although the performance is acutally not as good as MLP.

Checklist

• I have followed the Contribution Guidelines and Code of Conduct
• I have commented my code following the van der Schaar Lab Styleguide
• I have labelled this PR with the relevant Type labels
• My changes are covered by tests

[bcebere]

LGTM!
@robsdavis please review as well.

[robsdavis]

Thanks for you work on this @tztsai! I'll do my review and look at merging, if you can get the tests to pass again. Thanks so much!

[tztsai]

@bcebere @robsdavis There is an error when installing dependencies. Could you help check it?

[tztsai]

@bcebere @robsdavis There is an error when installing dependencies. Could you help check it?

I think the cause is the new Pip 23.1 released on April 15th. I kept its version to 23.0.1 and now it works fine.

[robsdavis]

Thanks @tztsai for fixing that. Simply pinning the version of pip is not normally something I would like to do, but we have already pinned the version for a couple of dependencies (and have issues logged to remove these constraints). So, we can look at removing this constraint on pip at the same time.