How to build coxph models with xgboost?

#1

I built coxph models with xgboost in R but got odd results.

For that model, I specified ‘objective’ as ‘survival:cox’ and ‘eval_metric’ as ‘cox-nloglik’; however, when I used the model to score a test dataset (new.score= predict(xgboost.cox, newdata = data.oot)), the results returned had nothing but ‘NaN’. There is no missing value in the test data and all variables used by the model are present.

I googled this error but it turns out that there is few example on this topic. Any suggestions?

Appreciate your help!

#2

In addition, I am also wondering whether the prediction of xgboost coxph returns the hazard and survival probability of each record at every unique time window in the training data, like what models built by coxph() do.

Thank you!