I’m doing a regression on a dataset using lightgbm. For the training data the response variable has a non normal distribution which is multimodal. However, the predictions out-of-fold have are normally distributed and have a single mode. Is there an assumption of normality made by light gbm ? Is there a way to make predictions which better match the required distribution?

  • $\begingroup$ tree based algorithms make no assumptions about variable distributions; I'd say your features are not able to separate the different modes. $\endgroup$ – Manu Valdés Jan 3 at 11:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.