Thanks for adding metric, early_stopping_rounds and progress to boosted_trees_classifier.create

User 2568 | 3/10/2016, 11:50:28 PM

The latest release (v1.8.4) adds options for metric and earlystoppingrounds to boostedtreesclassifier.create and the training/validation score per iteration can be accessed though the model field 'progress'.

This is very much appreciated. Not having this was a major drawback when using GraphLab vs XGBoost in Kaggle competitions. Firstly, this saves me a lot of trying to find the optimum iterations, and I can now draw a cool curve of the train and validation scores vs iterations like this

I noticed these options have not been added to logisticclassifier.create. Are these options only for boostedtrees or will they be added to other classifiers/regressors over time?


User 954 | 3/11/2016, 12:02:33 AM

Hi Kevin,

Thanks for your feedback and we are happy that you find it useful. Currently these features are for boostedtreesclassifier and regression only. Enabling these features on other classifiers is on our roadmap.