[Documentation] Comparative benchmarks

[arthurmensch]

It would be great to provide scripts that compares various estimators on problems of different size. I reckon the most pedagogical way to do so would be to display learning curves on a train and test set.