Multi-Agent Experiments with simple-es

Note: This is a fork of simple-es

Usage

For installation and general usage notes, please visit simple-es.

Testing

If you want to check if everything works fine and all environment configurations can be run, execute:

./scripts/test.sh

Docker

If you want to build and run the project in a docker container, execute the following commands:

Build the container:

docker build . -t "es-experiment:base"

Run a sample experiment:

docker run -e seed=0 es-experiment:base bash ./scripts/run.sh

Run interactively:

docker run -it es-experiment:base bash

Changes to simple-es

• adds dockerization
• add package version info to requirements.txt
• adds multi-agent support for gym environments
• uses a new neural network model with a hidden dimension of 64 and the reLU activation function
• adds logs for mean return and mean steps per episode that are saved to file
• adds the following multi-agent environments:
  • Level Based Foraging (lbforaging)
  • Robotic Warehouse (rware)
  • Multi-Particle Environments (MPE):
    • speaker-listener
    • adversary
• bug fixes:
  • fix NoneType error when passing the argument --save-gif to test.py
  • the pettingzoo environments now use the max_step configuration defined in the environment config yaml-files. (This did not work properly before, it always run just 1 step per episode)