Improve the ratio explanation in the docstring of the samplers

[amueller]

http://contrib.scikit-learn.org/imbalanced-learn/dev/generated/imblearn.ensemble.BalancedBaggingClassifier.html#imblearn.ensemble.BalancedBaggingClassifier

'auto': correspond to 'all' with for over-sampling methods and 'not minority' for under-sampling methods. The classes targeted will be over-sampled or under-sampled to achieve an equal number of sample with the majority or minority class.

This is always undersampling in BalancedBaggingClassifier, right? So auto is always 'not minority'? If so, I feel this docstring can be simplified.

[glemaitre]

True. This is copy paste from other sampler. We should probably address that in all samplers.
Thanks.

[glemaitre]

Having some feedback it seems that our docstring is not good enough regarding the ratio.
I think that we should avoid to mention over-/under- sampling option whenever it is not useful and also, we should link to the example in the documentation.

Better explanation regarding the possibility to give a dict should be explained.

[glemaitre]

@amueller I would appreciate some input regarding the ratio argument.

Naming

ratio does not apply anymore. It was true at the 0.0.1 of imblearn but we support multi classes since then. Do you think it is worth to rename it to something more related to what it does. Any input is welcomed. ping @jorisvandenbossche since it does not like the naming ;)

API

Right now, we can pass some string, dict and callable. We decided to do so to not have 2 arguments (targeted classes and #samples to get) avoid tedious sanity check between those arguments (length, order, etc.). Right now, we have an issue with the cleaning under-sampling method which discard the number of samples to select and just considered the class. This is bringing some confusion to users.

So if you see a way to make this parameter easier to manipulate and intuitive for the end-user, it would be great.

[glemaitre]

Regarding API I moved the discussion #411