Same Test Data/ Same Model/ Different Results

#1

I’ve trained a model using xgboost on python.
When feeding it the same test feature I get different prediction probabilities. Why?

#2

Hey, what are your XGBoost parameters ? if you set col_sample and col_subsample = 1, the predicted probabilities don’t change. You can read up more about the parameters to understand this better.

#3

Thanks for your answer @Anjala-ar , unfortunately applying your suggestion didn’t solve the problem.

I’m more confused as to why the model design would allow such behavior in the first place.

My parameters are:

hyperparams = {‘base_score’: 0.5,
‘booster’: ‘dart’,
‘colsample_bylevel’: 1,
‘colsample_bynode’: 1,
‘colsample_bytree’: 1.0,
‘gamma’: 10.0,
‘learning_rate’: 0.1,
‘max_delta_step’: 0,
‘max_depth’: 3,
‘min_child_weight’: 3,
‘n_estimators’: 100,
‘n_jobs’: 1,
‘nthread’: None,
‘objective’: ‘binary:logistic’,
‘reg_alpha’: 0,
‘reg_lambda’: 0,
‘scale_pos_weight’: 1,
‘random_state’: 0,
‘seed’: 0}

#4

Can you use gbtree instead of Dart?

1 Like
#5

Thanks @hcho3 that solved the problem.