Allow transformers to work when no suitable variables are found

[solegalli]

Our transformers were originally designed to automatically select variables based on type. If the user passes a variable list, transformers will corroborate the type of those variables is suitable for the transformer.

If no variables of the suitable type are found, then the transformer raises an error and does not run.

Some users asked to allow transformers to continue even if they will not modify the dataframe. That means, allowing the transformers to "work" when the argument variables_ is an empty list. Instead of returning an error, the transformer now returns and empty list and it proceeds with fit and transform as usual, but basically both methods will do nothing.

I can see the benefit of this change: It enables users to create pipelines that can be used across datasets without breaking (and some specific examples highlighted in the issues below).

The downside: unaware users will think that they are modifying variables when they are not. It becomes more of a black box.

Thoughts?

Tagging you guys to gather your opinion if you have time :)

@glevv @ClaudioSalvatoreArcidiacono @ankitlade12 @david-cortes @sTomerG @MatheusHam

Related issues:
#567 #566 #599

[ClaudioSalvatoreArcidiacono]

Hey Sole, I think it could make sense to raise a warning instead of an exception, this way we could get best of both worlds, wdyt?

[ankitlade12]

I agree with Claudio—a UserWarning is the perfect middle ground.

While hard errors are useful during development to catch configuration mistakes, they often become a bottleneck in production pipelines where data schemas can vary slightly between batches. Allowing a "no-op" transition increases the robustness of universal pipelines.

This approach makes the transformers behave more like 'filters'—if no data matches the criteria, the pipeline continues smoothly but with a clear audit trail in the logs.

[david-cortes]

I stand by my earlier comment that I'd find this useful.

If a warning is thrown, would be ideal to make it a custom warning so that it'd be easier to filter it out.

[glevv]

Custom warning sounds good, but this will be a big change in the library

If we will go with this option, I also think that we should add more to user guide and/or documentation about the "filter" behaviour and downsides that it may cause (work in one instances, but not in others, basically conditional nature of the transformers).