Box 'Em All - An Approach to Solve Dots & Boxes with Reinforcement Learning

Overview

Box 'Em All is a project that aims to solve the classic game of Dots & Boxes using reinforcement learning techniques. The game involves two players taking turns to draw lines between dots on a grid. The objective is to complete more boxes than the opponent. This project implements various strategies, including Q-learning, to train agents to play the game effectively.

Game Rules

1. The game is played on a grid of dots.
2. Players take turns to draw a line between two adjacent dots, either horizontally or vertically.
3. When a player completes the fourth side of a box, they claim that box and earn a point.
4. The player who completes a box gets another turn.
5. The game ends when all boxes are completed.
6. The player with the most boxes at the end of the game wins.

Project

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/box-em-all.git
   cd box-em-all

2. Install dependencies:

   Make sure you have Poetry installed. Then run:

   poetry install

3. Activate the virtual environment:

   poetry shell

Training the Model

To train the Q-learning or DQN model, run main and ensure the right runtime:

python box-em-all/main.py

Playing the Game inline:

You can play the game against the trained model by setting [is_human]to True in the main.py

python box-em-all/main.py

Using Weights & Biases:

If you want to log the training process using Weights & Biases, set use_wandb to True and make sure you are logged in to Weights & Biases:

wandb login
python box-em-all/main.py

Customization

• Board Size: You can change the board size by modifying the board_size
• Hyperparameters: Adjust the the hyperparameter for example like alpha, gamma, epsilon to tune the Q-learning algorithm.
• Model Saving/Loading: The model is saved and loaded from the model directory. The model folder is not checked in to git and is only for your local environment. if you want to use a saved model, copy the file to the player-modelfolder. If you want to make it available as frontent player, add an entry to the opponents list in box-em-all/api.py

Game Interface

You can use the flask api to play dots and boxes games. To start the api, run:

python box-em-all/main.py

Based on this api, there is a frontend app (Next.js). See in the folder dots-and-boxes-frontend for Details.