We need to address the following API issues raised by @jnothman:

• fit_sample is not a great name, as I have argued elsewhere ;)
• The fact that imblearn.pipeline.Pipeline has a method called sample that fits its constituents (using fit_sample) rings loud alarm bells. IMO, the sample method should not exist.
• fit_sample needs, eventually, to be able to update things other than X and y (e.g. weights).