ProjectBarks / connect4-leaderboard

A simple connect4 leaderboard

Connect 4 Leaderboard

A simple connect four leaderboard that runs the benchmarks in a dockerized container for security and standardization.

Screenshot

Setup & Running

Installation

1. Install docker for your specific system
2. Install postgres
3. Import table template into database (todo)
4. Update .env with relevant changes
5. Pull the relevant docker image intended or use the same image listed in .env
6. Activate and create a virtual environment
7. Install requiurements pip install -r requirements.txt

Execution

1. Running the server: ./run-server.sh
2. Running the tournament cron job ./run-tournament.sh

Base Code

class Game(object):
    pass


class MinimaxAgent(object):
    """Smart agent -- uses minimax to determine the best move"""
    
    def __init__(self, color: str):
        """Agents use either RED (R) or BLACK (B) chips."""
        pass

    def move(self, game: Game) -> Game:
        """Returns the best move using minimax as a new game instance"""
        # YOU FILL THIS IN
        pass

Todo

• Add Match Cache - When recomputing the matches check which users have increased their submission count and only redo matches for those players. This would not dramatically decrease computation but could be substantial in cases where only one or two players are modifying their code between execution.
• Containerize Players - Divide players into independent docker containers. It is possible to manipulate the shadow agent to falsify results and unfairly share time/memory. Separating containers would further equalize play.
• Dojo - Allow players who submit their code to run in a 1v1 fashion playing back the results in realtime with a dramatic effect. It could also permit benchmarking against competitive agents before a total re-computation.
• Hash Passcodes - The fact passcodes are not encrypted right now is a huge security risk. All passcodes should be encrypted and a bash script should be created to reset passcodes.
• Game Limits - There is the possibility for a player to capitalize on the time and memory against the other agent. To create fair play both players should be in their own docker instances with shared communication. The simplest current addition to the existing infrastructure would be to give an instant win to the agent who took less time. Cannot give an immediate win due to the halting problem.
• Duration Ordering - Agents with identical win rates should be ordered by least amount of time taken to win as a secondary sort component.
• Upload Files - Code must be submitted in plain text but it users should be allowed to upload files instead.
• Elo Rating System - A potential alternative to match cache is using the Elo rating system for rankings.

License

MIT License