Are you planning to add multi-output-regression capability to FLAML

[gmkyadav]

Current software support single output regression. It would be helpful if you can add multi-output-regression capability

[sonichi]

Thanks for the suggestion. Could you share what's the application and what learner is currently used? It will help us prioritize.

[sonichi]

Actually, I think you can simply inherit flaml.AutoML class, override its fit() function, and then use MultiOutputRegressor(ChildAutoML). Make sure class ChildAutoML meets the requirement of MultiOutputRegressor. For example:

class ChildAutoML(flaml.AutoML):

    def fit(X_train, y_train):
         super().fit(X_train, y_train, task="regression", time_budget=your_time_budget, **your_other_kwargs)

[gmkyadav]

Thanks for the suggestion. Will try and let you know.

[sonichi]

@gmkyadav Does it work? The same question is asked in #277

[sonichi]

With version v0.8.0, you don't need to create a new class. See an example at:

FLAML/test/automl/test_regression.py

Line 198 in 85e2186

def test_multioutput():

[sonichi]

See another example using RegressorChain: #306