ChaCha for Online AutoML

Abstract

We propose the ChaCha (Champion-Challengers) algorithm for making an online choice of hyperparameters in online learning settings. ChaCha handles the process of determining a champion and scheduling a set of `live' challengers over time based on sample complexity bounds. It is guaranteed to have sublinear regret after the optimal configuration is added into consideration by an application-dependent oracle based on the champions. Empirically, we show that ChaCha provides good performance across a wide array of datasets when optimizing over featurization and hyperparameter decisions.

arXiv comment16 pages (including supplementary appendix). Appearing at ICML 2021

References in corpus (7)

• Practical Bayesian Optimization of Machine Learning Algorithms
• OpenML: networked science in machine learning
• Scalable Global Optimization via Local Bayesian Optimization
• Autotune: A Derivative-free Optimization Framework for Hyperparameter Tuning
• Scalable Nonlinear Learning with Adaptive Polynomial Expansions
• Frugal Optimization for Cost-related Hyperparameters
• Model Selection for Production System via Automated Online Experiments