scikit-learn compatibility #11
Description
While I am trying to make our classes compatible with sklearn as much as possible (in #10 we start subclassing BaseEstimator), there are some limits to this at the moment. Notably, the signature of fit in sklearn estimator should be fit(X, y), while we have fit(X, y, geometry). I believe that what we have is the best option but it makes non-conformant.
We fail sklearn.utils.estimator_checks.check_estimator checks due to expected geometry and stuff like GridSearchCV does not work either as it is unable to pass geometry to predict. Same issue applies to score.
We could potentially pass geometry within init but then any CV splits would break anyway. Alternatively, geometry could be part of X but then none of the transformers in sklearn pipelines will work.
I am not sure if there is a way to make our API fully sklearn compatible at the moment.