EconML icon indicating copy to clipboard operation
EconML copied to clipboard

Published 20 hours ago • verified publisher icon py-why

new topic in discussion

Open ysong2123 opened this issue 3 years ago • 1 comments

Hi team,

is there any reason to adopt randomforest as model_final in ForestDRLearner? how about 'boosting-tree' based model_final, such as lightgbm?

Regards, Yan

ysong2123 avatar Mar 22 '22 04:03 ysong2123

If you only care about point estimates, then I think you could use a lightgbm regressor as the final_model for a DRLearner. The advantage of the RegressionForest used as the final model in ForestDRLearner is that it allows us to use a "bootstrap of little bags" to provide inference (confidence intervals, pvalues, etc.) on the results.

kbattocchi avatar Mar 22 '22 13:03 kbattocchi