Tutorial-LLM_GP

Simple implementation of Evolutionary Computation with LLMs. It is an extension of https://github.com/flexgp/pony_gp with addition of LLM based operators. The LLM based operators are based on https://arxiv.org/abs/2401.07102. It uses python3.

Install

Install requirements

# Create virtual environment. Here named `venv_tutorial_llm_gp`
python3 -m venv ~/.venvs/venv_tutorial_llm_gp
# Activate virtual environment. Here `venv_tutorial_llm_gp`
source ~/.venvs/venv_tutorial_llm_gp/bin/activate
# Install requirements
pip install -r ./requirements.txt

Install as a package e.g. pip install -e .

OpenAI API

Set the environment variable for OPENAI_API_KEY to your API key. E.g.

export OPENAI_API_KEY = "Some long and safe hash"

Run

Paths are relative the repository root.

Symbolic Regression

You can run tutorial.ipynb

Run GP

python main.py -f tests/configurations/tutorial_gp_sr.yml -o results_gp --algorithm=tutorial_gp

Run LLM-GP

python main.py -f tests/configurations/tutorial_llm_gp_sr.yml -o results_llm_gp --algorithm=tutorial_llm_gp_mu_xo

tutorial_llm_gp output

tutorial_llm_gp prints some information to stdout regarding settings and search progress for each iteration, see tutorial_llm_gp.py:print_stats.

The output files have each generation as a list element, and each individual separated by a ,. They are written to:

tutorial_llm_gp_*_fitness_values.json
tutorial_llm_gp_*_length_values.json
tutorial_llm_gp_*_size_values.json
tutorial_llm_gp_*_solution_values.json

Usage

usage: main.py [-h] -f CONFIGURATION_FILE [-o OUTPUT_DIR] [--algorithm {tutorial_gp,tutorial_llm_gp_mu_xo}]

Run Tutorial_LLM-GP

options:
  -h, --help            show this help message and exit
  -f CONFIGURATION_FILE, --configuration_file CONFIGURATION_FILE
                        YAML configuration file. E.g. configurations/tutorial_llm_gp_sr.yml
  -o OUTPUT_DIR, --output_dir OUTPUT_DIR
                        Path to directory for output files. E.g. tutorial_llm_gp_output
  --algorithm {tutorial_gp,tutorial_llm_gp_mu_xo}
                        Algorithms

Settings

Configurations are in .yml format, see examples in folder configurations.

Test

Tests are in tests folder. E.g run with

python -m unittest tests.test_main