Improve preprocessing and training for large datasets

[WhiteHoodHacker]

For some datasets containing either a large number of labels or a large amount of points, scikit fails to iterate through all the points to produce a complete trained model.

[WhiteHoodHacker]

Fixed in c65fe87