GOSS parameters in task-oriented AutoML

[ohenry-proda]

Hi,

I'm trying to tune a GOSS-enabled LGBMRegressor using the AutoML and LGBMEstimator from FLAML. This works great but it doesn't seem to tune the top_rate and other_rate associated with GOSS. Is there a way to enable this without wrapping the estimator in a function and tuning that?

[sonichi]

Have you considered the shortcut to override the search space? https://microsoft.github.io/FLAML/docs/Use-Cases/Task-Oriented-AutoML#a-shortcut-to-override-the-search-space