ax-platform

What it is and what it does

Ax is a machine-learning-guided optimization platform designed to automate the process of exploring parameter spaces and finding optimal configurations with minimal resource expenditure. It abstracts away complex optimization details by providing sensible defaults while remaining highly configurable for researchers and practitioners. The platform supports Bayesian optimization (via BoTorch) and bandit optimization, handles complex search spaces with multiple objectives, parameter constraints, and noisy observations, and can suggest multiple designs for parallel evaluation.

The package is built for production deployment with automation, orchestration, and robust error handling. It integrates with visualization tools (plotly, graphviz) and optional storage backends (MySQL via extras), making it suitable for both interactive experimentation in notebooks and large-scale automated optimization workflows. Its main dependencies are BoTorch for Bayesian optimization algorithms, pandas and scipy for data handling, scikit-learn for machine learning utilities, and various visualization and symbolic computation libraries.

Use it for:

• Hyperparameter tuning for machine learning models by iteratively exploring configurations and identifying optimal settings.
• Multi-objective optimization where you need to balance competing goals and constraints across a parameter space.
• Automated experiment design and execution in research or engineering where you want to minimize evaluation cost.
• A/B testing and online experimentation with adaptive allocation of resources to promising variants.
• Bandit-style exploration problems where you need to balance exploration and exploitation in real time.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Ax is a platform for adaptive experimentation that uses machine-learning-guided optimization to iteratively explore parameter spaces and identify optimal configurations. It supports Bayesian optimization and bandit optimization strategies powered by BoTorch.

Yes. Ax is actively maintained, has no known vulnerabilities, installs with low friction, and is licensed permissively (MIT). It is production-stable and suitable for anyone doing Bayesian or adaptive optimization. Install it if you need a general-purpose optimization platform; the main constraint is the Python 3.11+ requirement.

Install

ax-platform on PyPI

pip install ax-platform

uv add ax-platform

poetry add ax-platform

Installing ax-platform

Before you install

Low friction: pure Python wheel distribution. Active maintenance with recent releases; last commit 2026-08-13. Requires Python 3.11 or newer. Eleven runtime dependencies including botorch, scipy, scikit-learn, and visualization libraries (plotly, graphviz) are all standard packages.

License in practice

MIT license (permissive) means you can use, modify, and distribute Ax with minimal restrictions in commercial and open-source projects, provided you include the license notice.

Quickstart

pip install ax-platform

from ax import Client, RangeParameterConfig
from ax.utils.common import ParameterType

client = Client()
client.configure_experiment(
    parameters=[
        RangeParameterConfig(
            name="x",
            bounds=(-10.0, 10.0),
            parameter_type=ParameterType.FLOAT,
        )
    ]
)
client.configure_optimization(objective="my_metric")
for trial_index, params in client.get_next_trials(max_trials=1).items():
    client.complete_trial(trial_index=trial_index, raw_data={"my_metric": 0.5})

Requires Python 3.11 or newer.

Verify before relying

• Whether the notebook extra (ipywidgets integration) is required for Jupyter use or optional.
• Performance characteristics and scalability limits for large parameter spaces or many parallel trials.
• Whether fully_bayesian extra (JAX/NumPyro) is necessary for production use or only for specific model types.

Package facts

Evidence: ax_platform-1.3.1-py3-none-any.whl

Keywords: Experimentation, Optimization

Tags

More Artificial Intelligence packages

Further reading