Hi, I've been using the XGBRegressor with the enable_categoricals=True parameter and passing pd.Categoricals to that. When I use XGBFIR to generate the interaction importances, there are names such as: feature_1:{0,3,4,7,...}|feature_2 (or sometimes feature_2{0,1,...} if feature_2 is also categorical). Is this the splits that XGB used? (i.e. a tree with feature_1 as the first splitting variable, using the set {0,3,4,7,...} as the leaf. Then feature_2 as the 2nd split?)
Hi, I've been using the
XGBRegressorwith theenable_categoricals=Trueparameter and passingpd.Categoricals to that. When I use XGBFIR to generate the interaction importances, there are names such as:feature_1:{0,3,4,7,...}|feature_2(or sometimesfeature_2{0,1,...}iffeature_2is also categorical). Is this the splits that XGB used? (i.e. a tree withfeature_1as the first splitting variable, using the set{0,3,4,7,...}as the leaf. Thenfeature_2as the 2nd split?)