This repository contains unofficial code reproducing Agent57, which outperformed humans in all Atari games.
agent.py
define agent to play a supecific environment.
buffer.py
define buffer to store experiences with priorites.
learner.py
define learner to update parameter such as q networks and functions related to intrinsic reward.
main.py
run the main pipeline.
model.py
define some models such as q network and functions related to intrinsic reward.
segment_tree.py
define segment tree which decide segment index according to the priority.
tester.py
define tester which test performance of Agent57.
utils.py
define some classes and functions such as UCB and Retrace operator.
python==3.9.5
matplotlib==3.4.2
ray==1.4.1
lz4==3.1.3
numpy==1.21.0
omegaconf==2.1.1
torch==1.9.0
pip install -r requirements.txt
python main.py
Agent57: Outperforming the Atari Human Benchmark