Most discussions are focus on the features importance for all trees in XGBoost.
But I would like to get to know the features contribute for each output. Since each observation may has different paths. They would get the output because of different features. And different features may contribute to the output high or low.
In regression XGBoost may measure the contribution with R-square in my imagination. I’m not sure is it correct. And I’ve no idea how it looks like in classification XGBoost .
Is there any idea? Thank you.