Question about intercept and coefficients when usingXGBRegressor

ivan-marroquin · July 13, 2020, 4:10pm

Hi all,

I noticed that I can configure initial values for intercept and coefficient when using XGBRegressor - see below:

model= xgb.XGBRegressor(max_depth= 4, n_estimators= 50, verbosity= 0, tree_method= ‘exact’, n_jobs= self.cpus, booster= ‘gblinear’, objective= ‘reg:squarederror’, learning_rate= 0.10, gamma= 0.25, reg_alpha= 0.50, reg_lambda= 0.70, random_state= self.random_state[0], intercept_= pool_intercept, coef_= pool_coefficients)

My questions are the following: how does this command influence the way the regressor model is trained? is it normal to expect that the trained model will end with different intercept and coefficient values? Or, it will keep the same values?

Many thanks,

Ivan

hcho3 · July 16, 2020, 3:47am

XGBoost does not use intercept_ and coef_ arguments at all. So they will have no effect.

Only the arguments listed below will affect XGBoost training:

github.com

dmlc/xgboost/blob/ae18a094b065071612d0df0340e5aa8fa2c42482/python-package/xgboost/sklearn.py#L203-L213


def __init__(self, max_depth=None, learning_rate=None, n_estimators=100,
             verbosity=None, objective=None, booster=None,
             tree_method=None, n_jobs=None, gamma=None,
             min_child_weight=None, max_delta_step=None, subsample=None,
             colsample_bytree=None, colsample_bylevel=None,
             colsample_bynode=None, reg_alpha=None, reg_lambda=None,
             scale_pos_weight=None, base_score=None, random_state=None,
             missing=np.nan, num_parallel_tree=None,
             monotone_constraints=None, interaction_constraints=None,
             importance_type="gain", gpu_id=None,
             validate_parameters=None, **kwargs):

ivan-marroquin · July 17, 2020, 6:14pm

Thanks for the clarification