Hyperactive

Hyperactive is an open-source Python library for optimization and hyperparameter experiments. It provides a unified, experiment-based interface that cleanly separates optimization problems from the algorithms used to solve them. This design allows different optimization strategies to be swapped or compared without modifying experiment code.

The library combines native optimization algorithms with established frameworks such as Optuna and scikit-learn under a single, consistent API.

Purpose

Optimization workflows, especially hyperparameter tuning, are often fragmented across multiple tools and interfaces. Hyperactive reduces this complexity by offering a consistent abstraction that works equally well for research, experimentation, and production machine learning pipelines.

Core Features

• Unified API across multiple optimization backends

• Clear separation of what is optimized (experiments) and how it is optimized (algorithms)

• More than 20 optimization algorithms

• Support for mixed parameter spaces (categorical, integer, continuous)

• Direct integration with popular machine learning frameworks

• Stable and actively maintained since 2019

Optimization Backends

Gradient-Free-Optimizers

The native backend of Hyperactive, providing over 20 gradient-free optimization algorithms implemented from scratch. This backend is well suited for custom objective functions and research-oriented optimization tasks.

Supported methods include:

• Hill Climbing variants

• Simulated Annealing

• Particle Swarm Optimization

• Genetic Algorithms

• Bayesian Optimization

Optuna

Integration with the widely used Optuna framework, offering state-of-the-art samplers and pruning strategies for efficient hyperparameter optimization.

Supported techniques include:

• Tree-structured Parzen Estimator (TPE)

• CMA-ES for continuous search spaces

• Gaussian Process optimization

• Multi-objective optimization (e.g. NSGA-II / NSGA-III)

scikit-learn

Seamless access to familiar scikit-learn search strategies through Hyperactive’s experiment abstraction.

Supported methods include:

• GridSearchCV

• RandomizedSearchCV

• HalvingGridSearchCV

• HalvingRandomSearchCV

Experiment Abstraction

A core design principle of Hyperactive is the strict separation between:

• the definition of the optimization problem (objective function and search space)

• and the choice of the optimization algorithm

This enables easy comparison of algorithms and rapid iteration without changing the experimental setup.

Integrations

Hyperactive integrates directly with widely used machine learning frameworks.

scikit-learn

Hyperparameter tuning for estimators, including cross-validation workflows.

sktime

Optimization of time series forecasters and transformers.

skpro

Tuning of probabilistic regression models.

PyTorch

Optimization of neural network architectures and training parameters.

Use Cases

• Hyperparameter optimization for machine learning models

• Benchmarking and comparison of optimization algorithms

• Research and experimental optimization tasks

• Automated machine learning workflows

• Teaching and demonstration of optimization techniques

Open Source

Hyperactive is released under the MIT License and developed openly on GitHub. The project is fully documented, type-annotated, and tested, with active community involvement.

GC.OS supports Hyperactive as an open-source project that promotes transparent, reproducible, and extensible optimization workflows.