This repository contains an implementation of AGAC, as introduced in Adversarially Guided Actor-Critic (ICLR 2021).
This is the original TensorFlow implementation.
Find the PyTorch implementation here.
# create a new conda environment
conda create -n agac python=3.7
conda activate agac
# install dependencies
git clone git@github.com:yfletberliac/adversarially-guided-actor-critic.git
cd adversarially-guided-actor-critic
pip install -r requirements.txt
python run_minigrid.py
(follow Vizdoom install first)
python run_vizdoom.py
sudo apt-get install cmake libboost-all-dev libgtk2.0-dev libsdl2-dev python-numpy
pip install -e git://github.com/yfletberliac/vizdoomgym.git#egg=vizdoomgym
brew install cmake boost sdl2
pip install vizdoom==1.1.8
pip install pyglet==1.5.11 -e git://github.com/yfletberliac/vizdoomgym.git#egg=vizdoomgym
@inproceedings{
flet-berliac2021adversarially,
title={Adversarially Guided Actor-Critic},
author={Yannis Flet-Berliac and Johan Ferret and Olivier Pietquin and Philippe Preux and Matthieu Geist},
booktitle={International Conference on Learning Representations},
year={2021},
}
The code is an adaptation of Stable Baselines.
Thank you to @cibeah for the PyTorch implementation.