mae criterion is very slow compared to mse or friedman_mse for classification

fabsig / KTBoost

A Python package which implements several boosting algorithms with different combinations of base learners, optimization algorithms, and loss functions.

Other

63 stars 19 forks source link

mae criterion is very slow compared to mse or friedman_mse for classification #2

Closed flamby closed 5 years ago

flamby commented 5 years ago

Hello,

I did some bench w/ the 3 criterions available for classification, and mae is at least one order of magnitude slower. any reason?

model = KTBoost.BoostingClassifier(loss='deviance',update_step='hybrid', criterion='mae',
                                   n_estimators=50, random_state=seed)
model.fit(X_train,y_train)
probas = model.predict_proba(X_test)

fabsig commented 5 years ago

I guess that this is simply due to the fact that building regression trees with the 'mae' criterion is slower than with the other ones ('mse' or 'friedman_mse') for which there are fast version. I must admit though, that I am not an expert concerning regression trees with the MAE criterion.