Ax is an accessible, general-purpose platform for understanding, managing, deploying, and automating adaptive experiments.

Adaptive experimentation is the machine-learning guided process of iteratively exploring a (possibly infinite) parameter space in order to identify optimal configurations in a resource-efficient manner. Ax currently supports Bayesian optimization and bandit optimization as exploration strategies. Bayesian optimization in Ax is powered by BoTorch, a modern library for Bayesian optimization research built on PyTorch.

For full documentation and tutorials, see the Ax website [TODO: add link].

You need Python 3.6 or later to run Ax.

The required Python dependencies are:

• botorch
• jinja2
• pandas
• scipy
• simplejson
• sklearn
• plotly>=2.2.1,<3.0

NOTE: Both BoTorch and Ax are currently private repositories. This means that to download them, using pip, you need to make sure that you have an SSH key is registered with GitHub.

1. Install numpy and cython, if you don't already have them, since they are needed for parsing the setup.py files for Ax:

pip3 install cython numpy

2. Install PyTorch nightly build

See installation instructions via pip or conda here.

BoTorch uses advanced PyTorch features and thus currently requires PyTorch's nightly build (the requirement upon beta release will be PyTorch 1.1).

Recommendation for MacOS users: install PyTorch before installing BoTorch using the Anaconda package manager to get up to an order-of-magnitude speed-up for Bayesian optimization. At the moment, installing PyTorch from pip does not link against MKL, a library that optimizes mathematical computation for Intel processors.

3. Download and install BoTorch from the GitHub repo:

pip3 install git+ssh://git@github.com/pytorch/botorch.git

4. Download and install Ax from the GitHub repo:

pip3 install git+ssh://git@github.com/facebook/Ax.git

Install Ax via pip:

pip3 install ax-platform

Recommendation for MacOS users: install PyTorch before Ax by using the Anaconda package manager to get up to an order-of-magnitude speed-up for Bayesian optimization. At the moment, installing PyTorch from pip does not link against MKL, a library that optimizes mathematical computation for Intel processors.

Depending on your intended use of Ax, you may want to install Ax with optional dependencies.

If using Ax in Jupyter notebooks:

pip3 install git+ssh://git@github.com/facebook/Ax.git#egg=Ax[notebook]

If storing Ax experiments via SQLAlchemy in MySQL or SQLite:

pip3 install git+ssh://git@github.com/facebook/Ax.git#egg=Ax[mysql]

Note that instead of installation from Git, you can also clone a local version of the repo and then pip install with desired flags from the root of the local repo, e.g.:

pip3 install -e .[mysql]

To run a simple optimization loop in Ax (using the Booth response surface as the artificial evaluation function):

>>> from ax import optimize
>>> optimize(
        parameters=[
          {
            "name": "x1",
            "type": "range",
            "bounds": [-10.0, 10.0],
          },
          {
            "name": "x2",
            "type": "range",
            "bounds": [-10.0, 10.0],
          },
        ],
        # Booth function
        evaluation_function=lambda p: p["x1"] + 2*p["x2"] - 7)**2 + (2*p["x1"] + p["x2"] - 5,
        minimize=True,
    )

{'x1': 1.02, 'x2': 2.97}  # global min is (1, 3)

See the CONTRIBUTING file for how to help out. You will also need to install the dependencies needed for development, which are listed in DEV_REQUIRES in setup.py, as follows:

pip3 install git+ssh://git@github.com/facebook/Ax.git#egg=Ax[dev]

Ax is licensed under the MIT license.