DeepHyper¶
DeepHyper: Scalable Neural Architecture and Hyperparameter Search for Deep Neural Networks¶
DeepHyper is a scalable automated machine learning (AutoML) package for developing deep neural networks for scientific applications. It comprises two components:
Neural architecture search (NAS): It is designed for automatically searching for high-performing the deep neural network search_space.Hyperparameter search (HPS): It is designed for automatically searching for high-performing hyperparameters for a given deep neural network search_space.
DeepHyper provides an infrastructure that targets experimental research in NAS and HPS methods, scalability, and portability across diverse supercomputers. It comprises three modules:
benchmark: Set of test problems for NAS and HPS that can be used for comparing different search methods. They can serve as examples to build new user-defined problems.evaluator: Set of objects to run NAS and HPS on different target systems (from laptop to supercomputers) covering different use cases (quick/light experiments on laptop for testing and development to large production runs on supercomputers).search: Set of search methods for NAS and HPS. It provides a modular way to define new search HPS and NAS search methods and submodules for implementing HPS and NAS.
DeepHyper installation requires Python 3.6.
Installation
Getting started
Running search
Hyperparameter Search
Neural Architecture Search
Software