search

NVIDIA / spark-xgboost-examples

XGBoost GPU accelerated on Spark example applications
Apache License 2.0
52 stars 22 forks source link

Results mismatch during XGBoost Pyspark to Python conversion #26

Open gnani4444 opened 4 years ago

gnani4444 commented 4 years ago

Hi

1) I have trained and saved the xgboost pyspark using model.booster.save_model(path). I loaded into xgboost python using xgb.booster(path) The predictions of the XGBoost pyspark and python are different and the accuracy was dropped by 2 to 5% on the same data.

2) I have trained a XGBoost Python model with same features (from the pyspark). The results are different from Pyspark. The accuracy was very low in python when compared to pyspark.

Can you help why this was happening in both the cases.

GaryShen2008 commented 3 years ago

@wbo4958 Any idea about this issue?

wbo4958 commented 3 years ago

@gnani4444 Sorry for the late response,Since the version you tested is far far away from DMLC HEAD, Could you re-test the nv-release-1.3.0 which is based on dmlc's 1.3.0 release.