GitHub - hyperopt/hyperopt: Distributed Asynchronous Hyperparameter Optimization in Python

Abstraction: Python library for serial and parallel hyperparameter search over complex spaces

Key points:

• Hyperopt optimizes over awkward search spaces including real-valued, discrete, and conditional dimensions
• Three implemented algorithms: Random Search, TPE (Tree of Parzen Estimators), and ATPE (Adaptive TPE)
• Designed to accommodate Gaussian process and regression tree Bayesian methods (not yet implemented)
• Parallelization via MongoDB (MongoTrials) or Apache Spark (SparkTrials)
• API uses hp.choice, hp.uniform, hp.lognormal to define search spaces; fmin with tpe.suggest to minimize
• Cited as: Bergstra et al., ICML 2013, "Making a Science of Model Search: Hyperparameter Optimization in Hundreds of Dimensions for Vision Architectures"

Connections: Hyperopt · Github · Hyperparameter Optimization · Bayesian Optimization · Automl

Source: https://github.com/hyperopt/hyperopt