Add a test for the following:

A sample weight of 2 (3, 4, 5, n) on a given sample should have the same effect as placing this sample 2 (3, 4, 5, n) times in the data. This should hold for many estimators, especially those that fully optimize a convex loss.