Followup to #190, we have an issue with using it in GridSearchCV. The hyperparameter optimizers expect that param_grid be a dict of {parameter: [ values ]}, but Incremental doesn't really take any parameters. We're really searching over the parameters of estimator. Some solutions

1. Enhance scikit-learn to allow . access to attributes, something like param_grid={'estimator.alpha': [0.1, 10]}
2. rewrite Incremental.set_params (and maybe get_params) to pass things through

I'm going to explore number 2 right now. Might send an email to the scikit-learn mailing list to see if other meta-estimators have run into this, though a quick search didn't turn anything up.