Is there a way to obtain the baseline predictions from an xgboost cox model?
Baseline cox prediction
I am afraid not. However, as suggested by Peter Strom, there might be some workarounds. In short, you would need to fit the baseline hazard in another package and use the predicted output from XGBoost as multipliers to the fitted baseline. I haven’t tried it myself, though
I agree with the original post that it would be great if more examples could be provided.
@jyarmus @arturomoncadatorres This has to do with limitations of the Cox model, which only gives you the partial likelihood. We’d need to implement Breslow’s estimator to estimate cumulative baseline hazard.
Alternatively, you may want to choose Accelerated Failure Time model. We recently added support for it in XGBoost: https://xgboost.readthedocs.io/en/latest/tutorials/aft_survival_analysis.html