holland-gym

Evolving Neural Net Agents on OpenAI's Gym Environments using Holland.

Each environment tested has its own directory within the root directory of this project. Each of these environment directories will generally contain the following files:

• [env]_agent.py — contains the Agent class for the environment
• evolve.py — contains all code related to evolution of agents
• genomes.json — stores results from the best (or most recent) runs of evolution (storage parameters can be configured in evolve.py)
• display.py — displays a trial in the environment of the best individual from genomes.json

The root-level file agent.py contains the Agent class from which all specific environment Agents inherit.

Running Evolution

To evolve new individuals for a particular environment, simply run [env]/evolve.py. Note that you may want to configure the stoarge parameters in this file so that the tracked [env]/genomes.json is not overwritten.

Displaying Individuals

To run a trial (with rendering on) of the most fit individual stored in [env]/genomes.json, run [env]/display.py.