Hi!
I was reading the docs trying to understand the output when we set objective:aft
for survival modeling with xgboost. There is also this post that says that T(x) is the prediction of predict.xgb.Booster pretty much. If that is true, then the output values are not survival times since they need some extra calculations in there to compute that, the σΖ part. In the paper, the authors write though: “Although the XGBoost predict method only computes a point estimate (mean) of the survival time for each individual in the population, …”.
If not, then the prediction values must be somekind of linear predictions (lp) or exp(lp) as is the case with Cox (which is mentioned in the doc)? Which one is it is my question
BR, John.