GPyOpt

GPyOpt is a Python-based Bayesian optimization library built on Gaussian processes to efficiently solve expensive black-box tuning problems.

GPyOpt leverages the GPy framework to automate global optimization for functions where evaluations are computationally expensive (such as hyperparameter tuning for deep learning). It supports a wide range of acquisition functions: Expected Improvement (EI), Probability of Improvement (PI), and Lower Confidence Bound (LCB). The library manages complex input spaces including continuous, discrete, and categorical variables. By utilizing batch evaluations and parallel processing, GPyOpt helps researchers find optimal configurations with fewer iterations than standard grid or random search methods.

https://github.com/SheffieldML/GPyOpt