Description

I believe there is an issue with the coef_ description. The method is presented as :
(1 / (2 * n_samples)) * ||Y - XW||^2_Fro + alpha * ||W||_21
but coef_ returns W.T, currently : W.shape= [n_features,n_tasks] but coef_.shape=[n_tasks,n_features], cf. coordinate_descent.py--L1612

Why not outputting W directly? Instead of transposing the output, at least the doc should be consistent with the current implementation.
I am sending a PR for the second solution.

@agramfort