|
--- |
|
license: mit |
|
language: |
|
- en |
|
--- |
|
|
|
# README for Flappy-AI with Reinforcement Learning |
|
|
|
Model for the repository [flappy-ai](https://github.com/amc-madalin/flappy-ai), an project that utilizes reinforcement learning, to train an AI agent to master the game of Flappy Bird. This model is the 20th episode for the following hyperparameters: |
|
|
|
# File: config.yml |
|
``` |
|
game_screen: |
|
width: 800 |
|
height: 600 |
|
|
|
game_mechanics: |
|
gravity: 5 |
|
flap_strength: 20 |
|
pipe_speed: 2 |
|
pipe_width: 100 |
|
pipe_gap: 200 |
|
character_size: 50 |
|
collision_penalty: -1 # Penalty for hitting a pipe |
|
pass_reward: 1 # Reward for successfully passing a pipe |
|
height_penalty: -0.1 # Penalty for flying too high or too low |
|
|
|
initial_positions: |
|
character_x: 350 |
|
character_y: 250 |
|
pipe_x: 800 |
|
pipe_height_min: 100 |
|
pipe_height_max: 400 |
|
|
|
ai_config: |
|
epsilon: 0.3 |
|
epsilon_min: 0.01 |
|
epsilon_decay: 0.98 |
|
gamma: 0.1 |
|
learning_rate: 0.0001 |
|
batch_size: 32 |
|
buffer_size: 5000 |
|
num_episodes: 500 |
|
|
|
other_settings: |
|
random_seed: 42 |
|
|
|
model: |
|
save_path: ./experiments/ |
|
load_path: ~ |
|
``` |
|
## Project Structure |
|
|
|
- `ai/`: This directory houses the core AI components, including the Q-network for decision making, the replay buffer for learning from past actions, and the training logic essential for the agent's learning process. |
|
- `assets/`: Here, you'll find all the graphical and audio assets utilized in the project, enriching the game's visual and auditory experience. |
|
- `configs/`: Contains various configuration files that allow customization and fine-tuning of the AI agent and the game environment. |
|
- `environment.yml`: A comprehensive list of dependencies required for the project, ensuring consistent setup across different environments. |
|
- `experiments/`: This section includes experimental code and data, showcasing the trials and tests conducted during the development process. |
|
- `game/`: Contains the game's logic and rendering code, the heart of the Flappy Bird game that the AI interacts with. |
|
- `model/`: Here lies the trained AI model, a testament to the learning and adaptation achieved through reinforcement learning. |
|
- `runs/`: Stores outputs from various training runs, providing insights into the AI's performance and learning progression. |
|
- `utils/`: A collection of utility scripts and modules that support various functionalities across the project. |
|
- `train.py`: The primary script used to kickstart the AI agent's training process. |
|
- `test_game.py`: A testing script to run the game independently of the AI agent, useful for debugging and game playthroughs. |
|
|
|
## Installation |
|
|
|
To set up this project on your local machine, follow these steps: |
|
|
|
1. Clone the repository: |
|
```sh |
|
git clone https://github.com/amc-madalin/flappy-ai.git |
|
``` |
|
2. Navigate to the project directory: |
|
```sh |
|
cd flappy-ai |
|
``` |
|
3. Install the required dependencies: |
|
```sh |
|
conda env create -f environment.yml |
|
``` |
|
|
|
## Usage |
|
|
|
Once the installation is complete, you can start exploring Flappy-AI. |
|
|
|
1. Activate the project environment: |
|
```sh |
|
conda activate flappy-ai |
|
``` |
|
2. To begin training the AI agent, execute: |
|
```sh |
|
python train.py |
|
``` |
|
3. To test the game mechanics without the AI agent, run: |
|
```sh |
|
python test_game.py |
|
``` |
|
|
|
## License |
|
|
|
This project is released under the MIT License, promoting open-source collaboration and sharing. |
|
|
|
|
