Best model not saved using ray tune for xgboost training

Ray Tune
2

I think you need to do something similar to this to have Tune take care of your checkpoint saved in a distributed setup.

xgb.train(
         config,
         train_set,
         evals=[(test_set, "eval")],
         verbose_eval=False,
         callbacks=[TuneReportCheckpointCallback(filename="model.xgb")],
     )

Notice the callbacks section here.

See Tuning XGBoost parameters — Ray 1.13.0

As a side note, we are migrating away from this API in favor of Ray AIR. Please also take a look here: Hyperparameter tuning with XGBoostTrainer — Ray 3.0.0.dev0

1 Like