[Plugin] Sklearn integration

[cosmicBboy]

Flyte is an ML- and data- aware orchestrator, it would make sense to support the most popular ML frameworks as first-class citizen types in the Flyte type system.

For sklearn, there's really only one base class, which is the BaseEstimator.

We'll need to follow the conventions set in the flytekit.extras.pytorch module, by doing the following:

1. the __init__.py file should handle the case where the user doesn't have sklearn and joblib installed.
2. implement a TypeTransformer for the BaseEstimator class:
   • flyte docs: https://docs.flyte.org/projects/cookbook/en/latest/auto/core/extend_flyte/custom_types.html
   • numpy type transformer: https://github.com/flyteorg/flytekit/blob/master/flytekit/types/numpy/ndarray.py
3. add tests here: https://github.com/flyteorg/flytekit/tree/master/tests/flytekit/unit/extras. Test out:
   • BaseEstimators
   • Transformers (e.g. https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.StandardScaler.html?highlight=standardscaler#sklearn.preprocessing.StandardScaler)
   • sklearn.pipeline.Pipeline

Resources

• serialize/deserialize sklearn estimators: https://scikit-learn.org/stable/model_persistence.html
• joblib docs: https://joblib.readthedocs.io/en/latest/

Open Questions

• joblib/sklearn will complain if the an estimator is serialized in some version x and deserialized in another version y, but this is just a user warning... how do we handle this?

[SmritiSatyanV]

Is this something I can take up? @cosmicBboy

[kumare3]

@cosmicBboy can we add a more detailed description- I think this is a great good first issue for open source contributors?

[cosmicBboy]

assigned @SmritiSatyanV to this, gonna fill out #2759 and #2753 for others to pick up.

@kumare3 as a side note, I think we should host OSS planning meetings (similar to UnionML) so that we can get better involvement.

[techytushar]

I can pick this up, I have already worked on adding TypeTransformer for Tensorflow Models here: flyteorg/flytekit#1241
this seems to be very similar to it