sample_weight is not passed to sklearn.metrics in cross validation routines. In some cases, this causes cross validation routines to return suboptimal hyper-parameter configurations based on the training distribution.

This has a loose relationship to #251