While working on #11646 , I'm looking at the ElasticNetCV class and I humbly think the doc could be improved a bit:

• It's not clear at first glance what's the difference between ElasticNetCV and ElasticNet + GridSearch + best_param_.

  • Will ElasticNetCV refit the model with the best parameters, or does it just find the best parameters (and it's up to the user to refit)?
  • From what I understand, the advantage of ElasticNetCV is that it can take advantage of warm starting the estimators?

• It's not obvious what a "regularization path" is (related to previous point I suppose)

• The n_alphas and alphas parameters are confusing because they're mutually exclusive but it's not mentioned explicitly. alphas seems to have priority over n_alpha.

• The alphas_ attribute advertises The grid of alphas used for fitting, for each l1_ratio, but in reality it's just the alphas parameter. So if n_alphas was set, I would except alphas_ to display the automatically chosen alphas, but it's actually set to None.

I'd be happy to submit a PR if that's OK, I just want to make sure that I'm not missing something.