A Python interface for the GEGELATI Reinforcement Learning library — a framework for genetic programming and reinforcement learning using Tangled Program Graphs (TPG).

The library works with gymnasium environments only. We plan to incorporate more possibilities in the future like classification.

This library is a work in progress and should not be used in production.

Installation

Option 1: Install from source

git clone --recurse-submodules https://github.com/beuve/Gegelapy.git
cd Gegelapy
pip install .

Note: The --recurse-submodules flag is required to include the GEGELATI library.

Option 2: Install directly from GitHub

pip install git+https://github.com/beuve/Gegelapy.git

Note: In the future, the library should be installable directly from PyPI.

Usage Example

import gegelapy

# Create a Gym environment wrapper
env = gegelapy.GymEnvironment("CartPole-v1")

# Select the instructions the TPG will have access to
instructions = [gegelapy.add, 
                gegelapy.sub, 
                gegelapy.mul, 
                gegelapy.div]

# Initialize a trainer with the environment
trainer = gegelapy.TPG(env, instructions, max_init_outgoing_edges = 2)

# Train for a few generations
trainer.train(10)

# Save the trained model
trainer.save("cartpole_best.tpg")