Simple implementations of Cartesian Genetic Programming (CGP) and Linear Genetic Programming (LGP) in JAX. The main goal is to interface CGP and LGP with Brax (an implementation of several Mujoco environments in JAX) to leverage parallelization on GPU and speed up the optimization.
Clone the current repo, create the conda environment for the project, and activate it.
conda env create -f environment.yml
conda activate gpax
You will need to login to Weights&Biases for your experiments to be tracked online:
wandb login
The project also uses a Telegram bot to send you notifications about the status of experiments.
For it to work you need a telegram/token.yaml file structured as follows:
token: abc
chat_id: 012More info on telegram bots and their creation is available at https://core.telegram.org/bots/tutorial.