Scikit Regressor Integration

[mschmill]

I would be great if this package were to integrate in with scikit-learn, identify itself as a regressor, and implement the associated APIs. This would unlock scikit workflows as well as packages that build on scikit, like Yellowbrick. Overall the package is great, this would be greater!

https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/base.py#L372

[cjekel]

Thanks! I agree that getting pwlf 100% compatible with scikit-learn would be a nice long term goal. I think I want to merge the multi-dimensional support first https://github.com/cjekel/piecewise_linear_fit_py/issues/52

The major difference right now between the api's is that currently pwlf copies the x and y data on initialization, while scikit-learn would only need this during a .fit(x, y). (And the option of n_segments is backwards...) There was good reason for this distinction when the library first came out, but that's no longer the case.

I'd probably want to create a scikit-learn style class, without breaking backwards compatibility of the existing class.

[cerlymarco]

This can be done automatically by Linear Trees.

Linear Trees differ from Decision Trees because they compute linear approximation (instead of constant ones) fitting simple Linear Models in the leaves.

A sklearn compatible implementation is available here

[joshdunnlime]

@cerlymarco Kind of...

Linear Trees, from my understanding, (amazing work nonetheless) don't honour the continuously-smooth constraint that MVPWLR offers. This is important in many contexts, not least because it will reduce overfitting. I'm not sure there is a way with Linear Trees that can guarentee this constraint though I hope I am wrong :)

Of course, there are many applications where this isn't necissary. There are also applications where this lack is desirable (steps in functions).