█████ █████
██ ███████████████ ███
██ ████████████ ███████████ ██
██ █████████████ ██████████████ ██
██ ██████████████ ███████████████ █
█ ██████████ █████ ████ ███████████ ██
██ ████████ ████ █ ██ ██ ██████ ███████ █
█ █████████ ███ █ █ ██ █ █████████ █
█ ██████████ █ █████ █████ █ ███████████ █
██ █████████████████████ ██ ██ █████████████████████ █
█ ██████ █████ █ █ █ ██ ██ █████ ██████ █
█ ███████ ██ █ ██ ███ ███ ██ █████ ███████ █
█ ███████████████ █ ███ █ █ ███ ████████████████ █
█ ██████████ ██ █ ███ ██ █ ██ ██████████ █
█ ████████████ ███████ ███ ███████████ ███████████ █
█ ███████████████████ █ ██ █████████████████ █
█ ███████████ ██ ██████ ███████ ██ ███████████ █
█ ███████████████ ███████ ██ ██████ ███████████████ █
██ ███████████████████████ ██ ████████████████████████ █
█ ██████████████████████ ████ ██████████████████ ██ ██
██ █████████████████████ ████ ██████████████████████ █
█ ██████████████████████ ██ ██████████████████████ █
█ ██████████████████████ ██ ██████████████████████ ██
█ █████████████████████ ████████████████████ ██
█ ████████████████ █████████████████ ██
██ ████ ██ █ ██ ██ █ ███████ █
██ ███ ██ ███ ██ ██ █ █████ ███ ██ ████ ██
██ ████████████ ██ ██ █████████ ██
██ ████ ████ █████ ████ █████ ██
███ ███ ██████████ ████ ███
█████ █████
███
██████ ██ ██████ ██
██ ██ █████ ██████ █████ ██ ██ ██ ████ ██ ██ ██ ████
██ ██ ██ ██ ██ ██ ██ ██████ ██ █ ██ █ ██ ██ ██ █ ██
██ ██ ██ ██ ██ ██ ██ ██ ████ ██ █ ██ ██ █ ██
█████ █████ ██████ █████ ██████ ██ ████ ██ ███ ████
An open-source framework that combines large language models with evolutionary algorithms to discover and optimize high-performing code solutions.
This is a companion repository to science-codeevolve, and contains the complete experimental setup and results for the CodeEvolve paper.
This repository provides:
- Experimental configurations for reproducing all results
- Raw experimental data from paper runs (
.pkl,.py,.txtfiles) - Analysis notebooks with visualizations and statistical tests
The benchmark problems themselves are implemented in the main science-codeevolve repository.
science-codeevolve-experiments/
├── experiments/ # Raw experimental results
├── notebooks/ # Analysis and visualization
│ ├── experiment_analysis.ipynb # Main analysis notebook
│ └── figs/ # Generated figures from paper
└── README.md
-
experiments/: Contains results from paper experiments including:- Solution histories (
.pyfiles) - Checkpoints (
.pklfiles) - Logs and metadata (
.txtfiles) - Multiple runs with different seeds/configurations
- Solution histories (
-
notebooks/: Jupyter notebooks for analysisexperiment_analysis.ipynb: Statistical analysis and comparisons
First, install the main CodeEvolve framework:
# Clone and install CodeEvolve framework
git clone https://github.com/inter-co/science-codeevolve.git
cd science-codeevolve
conda env create -f environment.yml
conda activate codeevolve
cd ..# Clone this experiments repository
git clone https://github.com/inter-co/science-codeevolve-experiments.git
cd science-codeevolve-experimentsSet your LLM API credentials as environment variables:
export API_KEY=your_api_key_here
export API_BASE=your_api_base_urlThis repository supports two distinct notions of reproducibility:
The folder experiments/ contains the raw artifacts used in the paper (checkpoints, histories, logs). The notebook(s) in notebooks/ analyze those artifacts to generate the plots and comparisons. Re-running the analysis should reproduce the reported figures/tables as long as your analysis environment is compatible.
Exact numerical reproduction of a full evolutionary run is not guaranteed when using hosted LLM APIs.
Why:
- Many commercial LLM providers do not support deterministic sampling or do not honor
seed. - Even when a provider accepts
seed, outputs can vary due to backend nondeterminism (load balancing, infrastructure-level randomness, model version rollouts).
This is not a limitation of CodeEvolve’s evolutionary framework: CodeEvolve is seedable for its internal stochastic decisions, and it forwards model seed to OpenAI-compatible endpoints when supported. The remaining nondeterminism comes from the LLM backbone/provider.
@article{assumpção2025codeevolveopensourceevolutionary,
title={CodeEvolve: An open source evolutionary coding agent for algorithm discovery and optimization},
author={Henrique Assumpção and Diego Ferreira and Leandro Campos and Fabricio Murai},
year={2025},
eprint={2510.14150},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2510.14150},
}Experiments are versioned to match the main repository:
- v0.1.0: Initial release, corresponds to v1 of CodeEvolve's paper and v0.1.0 of the main repo.
- v0.2.0: Current release, corresponds to v3 of CodeEvolve's paper and v0.2.0 of the main repo.
- v0.2.1: Current release, corresponds to v4 of CodeEvolve's paper and v0.2.0 of the main repo.
The authors thank Bruno Grossi for his continuous support during the development of this project. We thank Fernando Augusto and Tiago Machado for useful conversations about possible applications of CodeEvolve. We also thank the OpenEvolve community for their inspiration and discussion about evolutionary coding agents.
All software is licensed under the Apache License, Version 2.0 (Apache 2.0); you may not use this file except in compliance with the Apache 2.0 license. You may obtain a copy of the Apache 2.0 license at: https://www.apache.org/licenses/LICENSE-2.0.
This is not an official Inter product.