Multi-output (multi-target) problems

[steffiel]

Hi, is it possible to do something like the multi-output from scikit-learn in Julia? Cause my input matrix has 3 features, and I have 12 different outputs per example. I tried to train 12 separate trees, but this didn't work.

[ablaom]

Multitarget regression is not possible - for regression currently the target must have Float64 type.

[ablaom]

Copying post of @fipelle from another thread:

---

Also, it seems that loss is only available for classification trees - not regression trees.

Is it possible to repurpose the existing code for classification trees to run regression tasks? It would be convenient both for

regression tasks with one target and a custom loss, and

multi-target problems (the current implementation for regression trees does not allow for labels that are not Float64 - i.e., single targets).

[ablaom]

This is an question here and I have wondered before why classification and regression have separate implementations. I diff'ed the two tree.jl files and found quite a few differences. For example, in classification there are array pre-allocations based on the total number of target classes, something you won't want to do in regression.

My initial guess is that generalizing regression.jl would be more useful than re-purposing classification.jl.