OpenFunsearch is an open-source project aiming to contribute to the scientific community by simplifying the use of the Funsearch algorithm. This project provides a detailed setup guide and easy-to-use scripts for running and analyzing Funsearch experiments.
The Funsearch algorithm, originally developed by Google DeepMind, is a cutting-edge method for solving complex search problems. For more information on the original research, please refer to the Nature paper.
You can also find the original Funsearch repository on GitHub.
We are not the first to try and work on Funsearch, changes have been made my github user: jonppe at Github Our build mostly continues on his work.
- Simplified Setup: Follow the detailed setup guide to install and configure the necessary environment.
- Command-line Execution: Run Funsearch experiments easily from the command line using
run_funsearch.py
. - Comprehensive Analysis: Analyze the results of your experiments using the
run_analyzer.ipynb
Jupyter notebook. - New word design problem: DNA Word problem added, which has not been tried before by Google, to show additional problems can be run.
To get started with OpenFunsearch, please refer to the Setup Guide for detailed installation and setup instructions.
To run a Funsearch experiment, you can use the run_funsearch.py
script. This script can be executed from the command line and can be altered to take all needed things such as LLM, Problem, Variables etc.
python run_funsearch.py
After completing a run, you can analyze the results using the run_analyzer.ipynb Jupyter notebook. This notebook provides an exploratory data analysis to get a quick overview over the results. Please refer to the specific eval_datetime.txt for analysis on a specific run.
If you have any questions or need further assistance, please feel free to contact me! This project is written by MsC Computer Science students Remco Stuij and Peter van Driel at Leiden University. Any suggestions or improvements are welcome.
This is not an official product.