hyperSPACE:Automated Optimization of Complex Processing Pipelines for pySPACE

Torben Hansing, Mario Michael Krell, Frank Kirchner

In: BayesOpt2016 - Bayesian Optimization: Black-box Optimization and Beyond. NIPS Workshop on Bayesian Optimization (BayesOpt-2016) befindet sich NeurIPS 2016 December 10 Barcelona Spain n.n. 12/2016.


Even though more and more algorithms are introduced for optimizing hyperparameters and complex processing pipelines, it still remains a cumbersome task for domain experts. Grid search is still used in most cases despite its deficiencies. In this paper, we combine the optimization library Hyperopt with the signal processing and classification environment pySPACE to completely automatize the optimization process. Even though no preliminary knowledge is required, interfaces for domain and algorithm experts were added to accelerate the optimization. As a proof of concept, the new framework is applied to electroencephalographic data which requires exhaustive optimization of a complex processing pipeline.


Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence