ogrisel / pygbm

Experimental Gradient Boosting Machines in Python with numba.
MIT License
183 stars 32 forks source link

Initialize predictions to something better than just 0 #47

Closed NicolasHug closed 5 years ago

NicolasHug commented 6 years ago

Saving it here so we remember: instead of initializing y_pred to 0 we may want to do something more sensible like boosting from the average target, or even accept an init predictor.