Hyperband with SMAC auxiliary optimizer
This library is intended to be a zero-dependency Scala implementation of a hybrid Hyperbandit and SMAC algorithm inspired by Bayesian Optimized Hyperbandit (BOHB) as implemented in HPBandSter.
Simalarly as to in original BOHB paper, we execute a number of benchmarks as part of our CI suite, and compare a reference XG-boost-based XGHyperSMAC algorithm to regular Hyperband as well as random parameter search (2x).
This project presents an abstract configuration space which can be defined with great flexibility. Additionally, it is intended to be easy to substitute the XG-Boost SMAC algorithm with arbitrary configuration selection models.
Basic usage examples can be found in test benchmarks.