The night is dark and full of terrors. Two teams must fight off the darkness, collect resources, and advance through the ages. Daytime finds a desperate rush to gather and build the resources that can carry you through the impending night. Plan and expand carefully -- any city that fails to produce enough light will be consumed by darkness.
Welcome to the Lux AI Challenge Season 1!
The Lux AI Challenge is a competition where competitors design agents to tackle a multi-variable optimization, resource gathering, and allocation problem in a 1v1 scenario against other competitors. In addition to optimization, successful agents must be capable of analyzing their opponents and developing appropriate policies to get the upper hand.
To get started, go to our Getting Started section. The competition runs until December 6th 2021 and submissions are due at 11:59PM UTC on the competition page: https://www.kaggle.com/c/lux-ai-2021
Thanks to our sponsors QuantCo, J Ventures, and QAImera, we have a $10,000 prize pool along with many other non-monetary prizes this year! For more information see https://www.lux-ai.org/sponsors-2021
Make sure to join our community discord at https://discord.gg/aWJt3UAcgn to chat, strategize, and learn with other competitors! We will be posting announcements on the Kaggle Forums and on the discord.
This was built by the Lux AI Challenge team, using the Dimensions package.
Season 1 specifications can be found here: https://lux-ai.org/specs-2021. These detail how the game works and what rules your agent must abide by.
You will need Node.js version 12 or above. See installation instructions here, you can just download the recommended version.
The next parts detail the recommended setup to develop and compete your bot. For users who wish to use Python and Jupyter Notebooks / Kaggle Interactive Notebooks, feel free to skip this section and follow the tutorial notebook
Open up the command line, and install the competition design with
npm install -g @lux-ai/2021-challenge@latest
You may ignore any warnings that show up, those are harmless. To run a match from the command line (CLI), simply run
lux-ai-2021 path/to/botfile path/to/otherbotfile
and the match will run with some logging and store error logs and a replay in a new errorlogs folder and replays folder. Logs stored in the errorlogs will include all error output and anything printed to standard error by your agent. You can watch the replay stored in the replays folder using our visualizer. To watch the replay locally, follow instructions here https://github.com/Lux-AI-Challenge/LuxViewer2021/
For a full list of commands from the CLI, run
lux-ai-2021 --help
or go to the next section to see more instructions on how to use the command line tool, including generating stateful replays and running local leaderboards for evaluation. You may also run this all in docker using the cli.sh file in this repo, see instructions here.
Each programming language has a starter kit, you can find general API documentation here: https://github.com/Lux-AI-Challenge/Lux-Design-2021/tree/master/kits
The kits folder in this repository holds all of the available starter kits you can use to start competing and building an AI agent and show you how to get started with your language of choice and run a match with that bot can be found. We strongly recommend reading through the documentation for your language of choice in the links below
- Python
- Javascript
- Rust (maintained by @tye-singwa)
- C++
- Java
- Typescript
- Kotlin (maintained by @Tolsi)
There are also many community provided tools to help people build better bots, feel free to check those out and use whatever suits your needs
Want to use another language but it's not supported? Feel free to suggest that language to our issues or even better, create a starter kit for the community to use and make a PR to this repository. See our CONTRIBUTING.md document for more information on this.
To stay up to date on changes and updates to the competition and the engine, watch for announcements on the forums or the Discord. See https://github.com/Lux-AI-Challenge/Lux-Design-2021/blob/master/ChangeLog.md for a full change log.
The CLI tool has several options. For example, one option is the seed and to set a seed of 100 simply run
lux-ai-2021 --seed=100 path/to/botfile path/to/otherbotfile
which will run a match using seed 100.
You can tell the CLI tool whether to store the agent logs or match replays via --storeLogs, --storeReplay. Set these boolean options like so
# to set to true
lux-ai-2021 --statefulReplay
# to set to false
lux-ai-2021 --storeLogs=false
By default the tool will generate minimum, action-based, replays that are small in size and work in the visualizer but it does not have state information e.g. resources on the map in each turn. To generate stateful replays, set the --statefulReplay option to true. To convert a action-based replay to a stateful one, set the --convertToStateful option to true and pass the file to convert.
Choose where the replay file is stored at by setting --out=path/to/file.json
You can also change the logging levels by setting --loglevel=x for number x from 0 to 4. The default is 2 which will print to terminal all game warnings and errors.
You can run your own local leaderboard / tournament to evaluate several bots at once via
lux-ai-2021 --rankSystem="trueskill" --tournament path/to/agent1 path/to/agent2 path/to/agent3 path/to/agent4 ...
This will run a leaderboard ranked by trueskill and print results as a table to your console. Agents are auto matched with opponents with similar ratings. Recommended to add --storeReplay=false --storeLogs=false as letting this run for a long time will generate a lot of replays and log files.
See lux-ai-2021 --help for more options.
This tool matches the lux-ai-2021 exactly, but runs on Ubuntu 18.04, the target system that the competition servers use. Make sure to first install docker
To then use the lux-ai-2021 CLI tool, simply call bash cli.sh and it will accept the same exact arguments. On the first run, it will build a docker image and run a container in the background. Future runs will then be much faster. Moreover, this uses a bind mount, so you can edit files locally on your computer and they will be reflected in the docker container and vice versa.
The only caveat of this tool is that it has no access to files in directories above the current working directory (the output of the pwd command).
Moreover, this tool will not inherit the same installed python packages on your computer. To add packages, please download the Dockerfile from this repo into the same directory as the cli.sh file and add installation commands like so to the bottom of the file
RUN pip3 install <package_name>
After changing the Dockerfile, run bash cli.sh clean to clean the old docker stuff and then use cli.sh as usual.
This is a list of all community tools built by our community!
- Lux AI Python Gym: A OpenAI compliant gym that replicates the entire Lux AI 2021 design. About 45x faster than the official environment for python agents and makes it easier to try more ML / search heavy approaches to the competition, maintained by glmcdona
- Optics: A simple, fast, top-down-view visualizer by rooklift
- Command Line Visualizer: A streamlined tool to watch a replay by running a single command in the CLI
- Public Notebooks: A collection of all public notebooks (in python usually) where competitors share their ideas, code, and more. There are all kinds of solutions here, from RL, Imitation Learning, to rule based approaches.
See the guide on contributing
Original design for season 1 concevied by Bovard and Stone
UI/UX Design by Isa
With balance testing help from David
We would like to thank our 3 sponsors, QuantCo, J Ventures, and QAImera this year for allowing us to provide a prize pool and exciting opportunities to our competitors! For more information on them, check them out here: https://www.lux-ai.org/sponsors-2021