COMP-424 AI Final Project

Preface

This GitHub repository features an engaging Colosseum Survival AI environment, tailored to test and evolve the StudentAgent from a strategic board game to a dynamic survival scenario. Utilizing a sophisticated genetic algorithm, the repository facilitates the enhancement of the agent's decision-making and survival strategies in a competitive Colosseum setting. It serves as an innovative platform for demonstrating the adaptability and robustness of the AI agent in diverse and challenging environments. Perfect for AI enthusiasts and researchers keen on exploring advanced AI behavior in survival simulations.

Description

Student Agent

This repository contains the implementation of StudentAgent, an advanced AI agent designed for a strategic two-player board game where players move on a grid and place barriers to control space. The StudentAgent uses sophisticated game theory algorithms and heuristics to make intelligent decisions. At its core, the agent employs:

• alpha-beta pruning with a transposition table
• iterative deepening search strategies
• a genetic algorithm with crossover and mutation to train heuristic weights

to efficiently explore possible moves, optimizing its decisions within a limited time frame. A transposition table is utilized to cache and recall previously evaluated game states, significantly speeding up the decision process. The agent's behavior is further refined by several custom heuristic functions, each assessing the game state from different strategic perspectives such as expansion potential, aggression, positional advantage, and the ability to extend barriers effectively. These heuristics are weighted and combined to form a comprehensive evaluation of each move. StudentAgent is an example of applying complex algorithms and heuristics to create a competitive and adaptable AI for grid-based board games.

Genetic Algorithm

This repository hosts a genetic algorithm for optimizing the heuristic weights of StudentAgent, an AI agent for a strategic two-player board game. Key features include:

• Robust Evolutionary Approach: Fine-tunes decision-making against various enemy strategies.
• Customizable Parameters: Includes population size, mutation and crossover rates, and number of weights.
• Fitness Evaluation: Based on performance against diverse enemy agents.
• Evolutionary Operations: Utilizes two-point crossover, Gaussian mutation, and tournament selection.
• Diversity Maintenance: Periodically reintroduces successful individuals to the population.
• Detailed Logging: Tracks best weights and win percentages against opponents for each generation.
• Outcome: Produces a strategically versatile AI agent capable of effective competition.

Testing

This repository includes a simulator for running and testing a strategic two-player board game where AI agents, including the StudentAgent, compete against each other. The simulator can be run from an Integrated Development Environment (IDE) or directly from the terminal, offering flexibility for different development and testing scenarios.

Installation:

Before running the simulator, you need to install the necessary Python libraries. You can install these dependencies using the following command:

pip3 install matplotlib==3.5.1 numpy pytest==7.0.1 tqdm==4.62.3 click==8.0.4 deap

These libraries include

• matplotlib for plotting and visualizing game states
• numpy for numerical operations
• pytest for running tests
• tqdm for progress bars in the terminal
• click for creating command-line interfaces
• deap for evolutionary algorithms

Running the Simulator:

• From an IDE: Simply import the required modules, instantiate the Simulator class with desired arguments, and call the run() or autoplay() method based on your testing needs.
• From the Terminal: Use command-line arguments to customize the simulation. Here's an example command to run the simulator:

  python3 simulator.py --player_1 student_agent --player_2 random_agent --board_size 10 --autoplay

Command-Line Arguments:

• --player_1: Specifies the first player agent (default: "random_agent").
• --player_2: Specifies the second player agent (default: "random_agent").
• --board_size: Sets a fixed board size for the game.
• --board_size_min: In autoplay mode, sets the minimum board size (default: 6).
• --board_size_max: In autoplay mode, sets the maximum board size (default: 12).
• --display: Enables the display of the game board (default: False).
• --display_delay: Sets the delay between moves in display mode (default: 0.4 seconds).
• --display_save: Enables saving of the game state (default: False).
• --display_save_path: Sets the path for saving game plots (default: "plots/").
• --autoplay: Enables autoplay mode to run multiple simulations automatically (default: False).
• --autoplay_runs: Specifies the number of games to run in autoplay mode (default: 1000).

About

This is a class project for COMP 424, McGill University

License

MIT