I'll try to work on this one.

Thanks @ldirer

Is this issue still open? If not, I'd like to pick it up!

I haven't really worked on it so you are welcome to pick it up if you want!

You might want to have a look at the forest code from 0.14 which is similar to what should be done in the multi-output setting.

https://github.com/scikit-learn/scikit-learn/blob/0.14.X/sklearn/ensemble/forest.py

Yup! I've been cribbing a good amount from the forest.py code.

I think everything is just about done, but I do have a quick question: is there a notion of a "decision function" in a multi-output setting? I did a quick look and all the estimators implementing a decision_function method seem to only have single-dim outputs.

With multi-output data, I would go with a list of arrays. One array for each output.

Okay - I've submitted a pull request #3798

implemented in #9707