The scikit-learn API defines classes_ as part of the API for classifier.

A similar handy thing for regressor models, IMO, would be to know if it was fit on a single or multioutput target. Currently, some regressors expose the n_outputs_ parameter, but other not. One can infer from the intercept_ or coef_ the number of target for liner model.

So I'm wondering if we could extend the API by extending n_outputs_ for all regressors the same way we have classes_ for classifiers?

Note that the tags do not help here because they inform whether or not an estimator is supporting multioutput.