Hyperopt

Hyperopt automates hyperparameter tuning in Python using Tree-of-Parzen Estimators (TPE) to navigate complex, high-dimensional search spaces.

This library optimizes objective functions across awkward search spaces involving real-valued, discrete, and conditional dimensions. It leverages the TPE algorithm (and Adaptive TPE) to outperform standard grid or random search strategies. You can scale experiments across clusters using MongoDB or Spark: integration with Scikit-learn and XGBoost is standard. The engine handles the heavy lifting of model selection by intelligently sampling the most promising parameter sets to minimize loss.

https://github.com/hyperopt/hyperopt