automl/SMAC3

What it solves

SMAC3 provides a robust framework for Bayesian Optimization, specifically designed to help users find the most effective hyperparameter configurations for machine learning algorithms, datasets, and applications.

How it works

It combines Bayesian Optimization with an aggressive racing mechanism to efficiently determine which of two configurations performs better. The system uses surrogate models—by default, Random Forests from scikit-learn, though it also supports the pyrfr C++ implementation—to guide the search for optimal hyperparameters.

Who it’s for

It is intended for machine learning practitioners and researchers who need to automate the tuning of hyperparameters to improve model performance.

Highlights

• Versatile Optimization: Supports multi-objective, multi-fidelity, and multi-threading natively.
• Flexible Interface: Includes an ask-and-tell interface and the ability to resume runs from where they left off.
• Integration: Works with ConfigSpace for defining optimization environments and can be visualized using DeepCAVE.
• Performance: Leverages C++ based Random Forests for efficient surrogate modeling.