Online Gradient Boosting

Authors: Alina Beygelzimer, Elad Hazan, Satyen Kale, Haipeng Luo

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we conduct some proof-of-concept experiments which show that our online boosting algorithms do obtain performance improvements over di erent classes of base learners. All experiments were done on a collection of 14 publically available regression and classification datasets (described in Section D in the supplementary material) using squared loss. The following table reports the average and the median, over the datasets, relative improvement in squared loss over the respective base learner.
Researcher Affiliation Collaboration Alina Beygelzimer Yahoo Labs New York, NY 10036 beygel@yahoo-inc.com Elad Hazan Princeton University Princeton, NJ 08540 ehazan@cs.princeton.edu Satyen Kale Yahoo Labs New York, NY 10036 satyen@yahoo-inc.com Haipeng Luo Princeton University Princeton, NJ 08540 haipengl@cs.princeton.edu
Pseudocode Yes Algorithm 1 Online Gradient Boosting for span(F) Algorithm 2 Online Gradient Boosting for CH(F)
Open Source Code No Is it possible to boost in an online fashion in practice with real base learners? To study this question, we implemented and evaluated Algorithms 1 and 2 within the Vowpal Wabbit (VW) open source machine learning system [23]. [23] VW. URL https://github.com/John Langford/vowpal_wabbit/. The paper states that the algorithms were implemented *within* Vowpal Wabbit (an existing open-source system), but it does not state that the code for *their specific implementations* of Algorithms 1 and 2 is open-source or provided.
Open Datasets Yes All experiments were done on a collection of 14 publically available regression and classification datasets (described in Section D in the supplementary material) using squared loss.
Dataset Splits Yes Parameters were tuned based on progressive validation loss on half of the dataset; reported is propressive validation loss on the remaining half. Progressive validation is a standard online validation technique, where each training example is used for testing before it is used for updating the model [3].
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or cloud instance types used for running the experiments. It only states that experiments were conducted.
Software Dependencies No The paper states that the algorithms were 'implemented and evaluated [...] within the Vowpal Wabbit (VW) open source machine learning system [23]', but it does not specify a version number for Vowpal Wabbit or any other software dependencies.
Experiment Setup Yes The only parameters tuned were the learning rate and the number of weak learners, as well as the step size parameter for Algorithm 1.