NEAT implementation based on neat-python library
First install python (v 3.5 is recommended for OpenAI libraries) and install conda or miniconda. Miniconda is recommended as its smaller and useful if you have space requirements.
Set up a conda environment by following the Project dependencies repo
Install the organisation's custom gym gym-ple libraries into the conda environment. These libraries extend openai gym. If you are looking on running gym-ple games, then you will need to install the PyGame-Learning-Environment from here.
The main file is NEAT.py. Experiment logs are written into log directory.
Run this file and if you want to save the std outputs for clarity use following command:
python NEAT.py [Environment Id] > output.log 2&>1
| Enviornment id | Info |
|---|---|
| CartPole-v0 | Standard implementation of gym cartpole |
| CartPole-v1 | Standard implementation of gym cartpole |
| MountainCar-v0 | Standard implementation of gym mountain car problem |
| MountainCarExtraLong-v0 | Custom implementation of gym mountain car problem where the episode length is 999. |
Use any of these environment id as an argument.
The code uses the environment id to extract properties from the properties directory. The arrangement of the properties directory is
properties/<Environment id>/Config
properties/<Environemt id>/neatem_properties.ini
Adding your environment involves creating a new directory of your environment id and add the two properties files used by the algorithm