Define inclusion criterion in imbalanced-learn in the documentation

[glemaitre]

I think that we never publicly exposed what would be the method inclusion in imbalanced-learn.

IMO, I think that we don't have to be as conservative as in scikit-learn to include new methods. However, I think that it is really important to have good documentation such that a new user can choose the appropriate method.

IMO, I think that we can include any method. However, we need to have a continuous benchmark tracking performance and computation time.

I think that the documentation should always refer to the benchmark.

@chkoar WDYT?

[chkoar]

I think that we never publicly exposed what would be the method inclusion in imbalanced-learn.

We did that explicitly here.

[chkoar]

However, we need to have a continuous benchmark tracking performance and computation time.

IMHO performance is not needed. At least for our regressions. Time is the key.

[glemaitre]

We did that explicitly here.

Oh I have short memory :)

[chkoar]

Well, that's better than having a 32bit machine in 2019!