I don’t understand the documentation about the xgboost Gain value in feature importance. I am refering to R implementation. Suppose I have binary classification task, so objective is set to ‘binary:logistic’ and I use the ‘auc’ as eval_metric. I looked at the implementation codes and gain of single split is a described as ‘change in loss’. But what loss is it? Is it binary:logistic loss value or change in eval_metric?
