benchmarx 0.0.11

Tools for benchmarking optimization methods

Benchmarx: Benchmarking Optimization Methods with Jax

Benchmarx is a Python library that enables you to benchmark various optimization methods in a reproducible manner using the Jax library. It provides a flexible framework for comparing the performance of different optimization algorithms on a wide range of optimization problems.

Features

• Benchmarx allows you to easily benchmark optimization algorithms provided by the Jaxopt library as well as custom algorithms.
• Reproducibility is ensured by providing a comprehensive set of parameters to control the benchmarking process, including RNG states.
• Comes with a collection of standard optimization problems, making it simple to evaluate algorithms on diverse tasks.
• Supports tracking and visualization of key metrics such as function values, gradient norms(if possible) and solution trajectories.
• Generate customizable plots and reports to aid in analyzing and presenting benchmark results with plotly library.
• All the data from the experiments could be saved in the json or pandas format for the later analysis.

Installation

You can install Benchmarx using pip:

pip install benchmarx

Usage

To be done

Examples

Refer to the examples provided in the repository for details on how to use this package.

• Gradient Descent on Quadratic problem

• Gradient Descent with custom Line Search

• Your own custom method

• Your own custom metric and interactive chart

• Stochastic Gradient Descent on Quadratic Problem

• Quadratic Problem based on real data

• Neural Network training

Documentation

Check out the full documentation at benchmarx.fmin.xyz.

Contribution

Contributions are welcome! If you have any suggestions or issues, feel free to open an issue or submit a pull request.