bayex 0.3.0

[!NOTE] Bayex is currently a minimal, personally developed implementation that requires further development for broader application. If you're interested in engaging with Jax and enhancing Bayex, your contributions would be highly welcomed and appreciated.

Installation | Usage | Reference docs

Bayex is a lightweight Bayesian optimization library designed for efficiency and flexibility, leveraging the power of JAX for high-performance numerical computations. This library aims to provide an easy-to-use interface for optimizing expensive-to-evaluate functions through Gaussian Process (GP) models and various acquisition functions. Whether you're maximizing or minimizing your objective function, Bayex offers a simple yet powerful set of tools to guide your search for optimal parameters.

Installation

Bayex can be installed using PyPI via pip:

pip install bayex

Usage

Using Bayex is quite simple despite its low level approach:

import jax
import numpy as np
import bayex

def f(x):
    return -(1.4 - 3 * x) * np.sin(18 * x)

domain = {'x': bayex.domain.Real(0.0, 2.0)}
optimizer = bayex.Optimizer(domain=domain, maximize=True, acq='EI')

# Define some prior evaluations to initialise the GP.
params = {'x': [0.0, 0.5, 1.0]}
ys = [f(x) for x in params['x']]
opt_state = optimizer.init(ys, params)

# Sample new points using Jax PRNG approach.
ori_key = jax.random.key(42)
for step in range(50):
    key = jax.random.fold_in(ori_key, step)
    new_params = optimizer.sample(key, opt_state)
    y_new = f(**new_params)
    opt_state = optimizer.fit(opt_state, y_new, new_params)

with the results being saved at opt_state.best_params and opt_state.best_score.

Documentation

Available at https://bayex.readthedocs.io/en/latest.

License

Bayex is licensed under the MIT License. See the LICENSE file for more details.