benchmarking functions

A collection of benchmarking functions for use with optimization algorithms.

This package provides a collection of standard test functions used to benchmark the performance of optimization algorithms. Each function exposes its callable form along with metadata such as its known global (and, where applicable, local) minima/maxima and suggested input bounds, making it easy to score an optimizer’s results against ground truth.

The original 10 functions (Branin, Goldstein-Price, Hartmann 3D, Hartmann 6D, Rosenbrock, Shekel 5, Shekel 7, Shekel 10, Shubert, and Six-Hump Camel) match the set used to benchmark the CyRxnOpt optimizer and are the same functions used in the SNOBFIT paper for its own benchmarking.

This repository is intended to grow beyond that original set of 10 test problems. It currently also includes Beale, Booth, Bukin N.6, Eggholder, Himmelblau, Holder Table, Matyas, Schwefel, Sphere, Styblinski-Tang, and Three-Hump Camel, with more functions expected to be added over time.

Installation

Clone the repository and install it with pip:

git clone https://github.com/RxnRover/benchmarking.git
cd benchmarking
pip install .

Usage

Each benchmarking function is implemented as a class that can be called directly with a list of input coordinates. For example, to evaluate the Branin function and inspect its metadata:

from benchmarking.functions.branin import Branin

branin = Branin()

# Evaluate the function at a point.
result = branin([-3.14, 12.275])

# Metadata about the function: known minima/maxima, bounds, etc.
print(branin.metadata)

# Convenience accessors are also available.
print(branin.min)  # value of the global minimum
print(branin.bounds)  # suggested input bounds

Functions can also be looked up by name, which is useful when the specific function to benchmark against is chosen dynamically (e.g. from a config file or command-line argument):

from benchmarking.evaluate import evaluate
from benchmarking.function_data import function_data

result = evaluate("branin", [-3.14, 12.275])
metadata = function_data("branin")

The full list of valid function names is available in benchmarking.function_ids.function_ids.

Making Changes & Contributing

This project uses pre-commit, please make sure to install it before making any changes:

# After cloning the repository
pip install pre-commit
cd benchmarking
pre-commit install