XGBClassifier( tree_method='hist', objective='multi:softprob', num_class=3, n_estimators=50, max_depth=5, learning_rate=0.1, eval_metric='mlogloss', subsample=1.0, colsample_bytree=1.0, random_state=42, n_jobs=1, device='cpu' ) This is my XGBoost classifier model. I am running same script for training the model on Windows machine and Linux machine. But the trained models are having different boosters/tree structures. I have verified that the data being used for training is same (using pd.testing.assert_frame_equal). I have verified that xgboost==2.1.1 and scikit-learn==1.5.1 are same on both the OS. I have used fixed random_state, and ensured that all the parameters used for the xgboost classifier are same on both OS using- assert model1.get_params() == model2.get_params(). But the trained models (saved with .json) when compared using- assert model1.get_booster().get_dump() == model2.get_booster().get_dump(), gives error (comparison results into False), meaning that the booster/tree structure is different in both models. Since the data used for training and the model parameters are same, I expect that the resulting models should be same. Please guide me to have the identical models on both the systems. Continue reading...
