1. 程式人生 > >[P/M] One-hot encoding is BAD to Boosting

[P/M] One-hot encoding is BAD to Boosting

One-hot encoding is not required for tree-models like RF and boostings. Here I would say categorical variable do not benefit boostings but opposite. The main idea is decision-tree based models have way to deal with numerical variable,variables don’t have to be encoded categorical for algorithm to do with.On the other way, creating too much categorical variables sparsely will do harm to tree-models.