I train my xgb model in Jupyter and save it with joblib. I open the model in Spyder and retrain it with the same X_train and y_train that was used for the initial training and I get different feature importance.

With another retrain (again same X_train, y_train) in Spyder I get the same results as the first retrain.

Is it possible that I get different results in Spyder than in Jupyter?

Note: The predictions don't change

Note 2: I extract feature importance with model.get_booster().get_score(importance_type = 'total_gain')

Note 3: I saved environment from Jupyter and used it in Spyder.