为什么SHAP值不累加预测概率?

时间:2020-05-26 22:36:00

标签: xgboost r-caret xgbclassifier

我想知道下面的设置是否正确,以便使用CARET的xgbDART方法从模型生成SHAP值。在此特定情况下的问题是SHAP值rowSums()不会累加预测分数。任何方向都将非常有帮助。这是我的设置:

#model scores
set.seed(200)
pred_mod<- predict(xgb_05_v6$finalModel, 
                   newdata= as.matrix(X1),
                   ntreelimit = param_dart$nrounds, 
                   objective = "binary:logistic",
                   outputmargin = TRUE
                   ) 

#SHAP values
set.seed(200)
SHAP_contribution<- predict(xgb_05_v6$finalModel, 
                    newdata= as.matrix(X1),
                    ntreelimit = param_dart$nrounds, 
                    objective = "binary:logistic",
                    approxcontrib = FALSE,
                    predcontrib = TRUE

) 

0 个答案:

没有答案