LoLGym for COMS4995 Deep Learning Project

About

This repo contains code to train an agent to play league of legends using the PPO algorithm and can be found in examples/full_game_ppo.py. In this project, we built an AI capable of playing and winning in the game League of Legends, a popular multiplayer game created by Riot Games. We implement a model that uses the existing deep reinforcement learning practice of Proximal Policy Optimization (PPO) alongside an adversarial multi-agent architecture to outperform existing hard-coded bots and human players in a 1v1 game.

This repo is forked from https://github.com/MiscellaneousStuff/lolgym

Installation

Install .NET

wget -q https://packages.microsoft.com/config/ubuntu/18.04/packages-microsoft-prod.deb -O packages-microsoft-prod.deb
sudo dpkg -i packages-microsoft-prod.deb
add-apt-repository universe
apt-get update
apt-get install apt-transport-https
apt-get update
apt-get install dotnet-sdk-3.1

Install the runtime

sudo apt-get update; \
  sudo apt-get install -y apt-transport-https && \
  sudo apt-get update && \
  sudo apt-get install -y dotnet-sdk-3.1

Clone GameServer

Download the 4.20 version of League game client

git clone https://github.com/MiscellaneousStuff/LeagueSandbox-RL-Learning
cd LeagueSandbox-RL-Learning && git checkout master && git branch && git submodule init && git submodule update

Build GameServer

cd LeagueSandbox-RL-Learning  && dotnet build .

Run game server to generate configs

cd /content/LeagueSandbox-RL-Learning/GameServerConsole/bin/Debug/netcoreapp3.0/ && \
/content/LeagueSandbox-RL-Learning/GameServerConsole/bin/Debug/netcoreapp3.0/GameServerConsole --redis_port 6379

Install PyLOL

git clone -v https://github.com/jjlee0802cu/pylol.git
pip3 install --upgrade pylol/

Install Redis Client and Redis Server

pip3 install redis
sudo apt-get install redis-server

Write config dirs (Game Server)

touch PATH/TO/config_dirs.txt
printf "[dirs]\ngameserver = PATH/TO/LeagueSandbox-RL-Learning/GameServerConsole/bin/Debug/netcoreapp3.0/\n" > PATH/TO/config_dirs.txt

Download LoL client

!wget --no-check-certificate --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate 'https://docs.google.com/uc?export=download&id=1vkrdzSxTN6FPP7A9R65bIEBNefU7vDJL' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=1vkrdzSxTN6FPP7A9R65bIEBNefU7vDJL" -O league-of-legends-420.tar.gz && rm -rf /tmp/cookies.txtInstall Winetricks

Install Wine

sudo dpkg --add-architecture i386
sudo apt update -y
sudo apt install wine64 wine32 -y

Install Winetricks

Without this you will get into the game, but your screen will be black.

sudo apt-get install winetricks -y
winetricks d3dx9

Extract LoL Client

tar -xzvf league-of-legends-420.tar.gz

Run Client

pip install pyautogui
pip install Xlib

Write config dirs (LoL Client)

touch PATH/TO/config_dirs.txt
printf "lolclient = /content/League-of-Legends-4-20/RADS/solutions/lol_game_client_sln/releases/0.0.1.68/deploy/" > config_dirs.txt

Install LolGym

git clone https://github.com/jjlee0802cu/lolgym.git
pip3 install -e lolgym/

Usage

PPO Agent trained against hard-coded scripted intermediate bot (Milestone)

Use commit 0c4958342468c1673e1f86046114b629f0268328

python3 ./lolgym/examples/full_game_ppo.py --epochs 200 --host <public_ip> --config_path "PATH/TO/config_dirs.txt" --run_client

Two PPO Agents trained against each other using adversarial multi-agent architecture (Final report)

Use commit 5031243750aaeca52dd4f2d310681a90554cdeda

python3 ./lolgym/examples/full_game_ppo.py --epochs 200 --host <public_ip> --config_path "PATH/TO/config_dirs.txt" --run_client