Suika Game Reinforcement Learning

This project implements a Reinforcement Learning (RL) agent to play the Suika Game (Watermelon Game) using Deep Q-Networks (DQN).

Quick Start

1. Install Dependencies

Make sure you have Python 3.10 environment. Then install the required packages:

pip install -r requirements.txt

2. Run the Trained Agent

We have provided a pre-trained model: suika_dqn_mlp_final.zip. To watch it play:

python rl_env/test_model.py --model suika_dqn_mlp_final.zip

Options:

• --episodes N: Run for N episodes (default: 1).
• --fps N: Limit playback speed to N FPS (default: 60).
• --stochastic: Use random actions based on probabilities (default: deterministic/best action).
• --empty: Start with an empty board (no initial random fruits).

Example:

python rl_env/test_model.py --model suika_dqn_mlp_final.zip --episodes 3 --fps 120

Other Ways to Run

Play the Game Yourself

If you want to play the game manually using your mouse:

python rl_env/human_play.py

• Controls: Move mouse to position the cloud, click to drop the fruit.

Train a New Agent

To train a fresh agent or continue training:

python rl_env/train.py

This will save checkpoints to models_dqn/ and logs to logs_dqn/ in the current directory.

File Overview

• suika_dqn_mlp_final.zip: The final trained DQN model ready for testing.
• rl_env/: Contains the Reinforcement Learning scripts.
  • test_model.py: Script to load and watch a trained model play.
  • train.py: Script to train the DQN agent.
  • human_play.py: Script for human gameplay.
  • suika_env.py: The Gymnasium environment wrapper for the game.
• suika/: Contains the core game logic and assets. Taken from an open source project seen here: https://github.com/Ole-Batting/suika
• requirements.txt: List of Python dependencies.