lightgbm.cv strange results in Random Forest mode

[candalfigomoro]

This is basically a continuation of this closed issue: #3387

With lightgbm's cv method, if I use 300 trees in Random Forest mode, I get 300 scores.

In comparison, when I use xgboost's cv method in Random Forest mode, I get just 1 score (this naturally arises from the fact that in xgboost we have decoupled parameters for boosted/parallel trees, see #3128).

What is the intepretation of the 300 scores? Is the 1st score the one I'd get with 1 tree and the 300th score the one I get with a Random Forest of 300 trees?

While for boosting iterations this kind of output makes sense, I think the interpretation of lightgbm's cv results in Random Forest mode is a bit confusing.

[shiyu1994]

@candalfigomoro Thanks for using LightGBM.

What is the intepretation of the 300 scores? Is the 1st score the one I'd get with 1 tree and the 300th score the one I get with a Random Forest of 300 trees?

Yes, that's correct. The ith value in the list represents the averaged mean or std across the folds using the first i iterations.
This is because rf is treated as a special type of boosting in LightGBM, which does not update the gradients between iterations.

[candalfigomoro]

@shiyu1994 Thank you!

[github-actions]

This issue has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.