sklean style api

[mariusvniekerk]

Initial draft of an sklearn style api

[amueller]

Actually, scikit-learn is explicitly designed in a way that you don't need to depend on it to satisfy it's API (you need to implement set_params and get_params, though that is very easy).

[amueller]

If you decide you want to depend on scikit-learn, I'd encourage you to run this function in your unit tests: https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/utils/estimator_checks.py#L219

It is new in 0.17 and checks scikit-learn compatibility.
Also see the docs here:
http://scikit-learn.org/dev/developers/contributing.html#rolling-your-own-estimator

(they also contain more details on set_params and get_params)

You could also make scikit-learn an optional dependency, so that it can be used to test compatibility but is not required by the user.

[ericdill]

@mariusvniekerk what's the status of this PR?

[mariusvniekerk]

Basically needs refactoring to be closer to Andreas' suggestion