DOC - Improving the user guide section about the tabular pipeline

[rcap107]

The user guide section on the tabular pipeline should be expanded with some additional examples on how to use the tabular pipeline for situations that are slightly more complex than what is currently explained.

• Show that the resulting pipeline can be modified by adding more steps
• Show how to use more specific transformers/steps
• Show that the model can be a pipeline that adds the steps needed by the user

[kudos07]

Hello @rcap107, Would like to take up this as well.

[rcap107]

Hi @kudos07, feel free to work on this, thanks

[rcap107]

For some additional detail: we would like to have an example that relies on the .steps attribute of the resulting pipeline and how that can be used with scikit-learn's make_pipeline to build a new pipeline that has more than just the steps in the tabular pipeline. Unfortunately I don't have a great example to give you for this off the top of my head.

It would also be useful to have a section on how to move from the tabular pipeline to a custom-made pipeline where we concatenate some of the steps of the tabular pipeline, but some of the parameters are changed manually. For example, we could show how to set the encoder for high cardinality features to a TextEncoder (though we might not be able to fit it in the doctests 🤔 ), or how we can combine ApplyToCols and the TableVectorizer to transform only a subset of columns.

Something else to explain is that the logic that is used by the tabular_pipeline is in fact quite simple, so users don't miss on a lot if they decide to write their own pipeline instead.