/Neural-Networks

This project implements a simple neural network-based agent to play a Flappy Bird-inspired game environment. The goal is to train multiple agents to navigate through the environment and achieve the highest score possible by avoiding obstacles.

Primary LanguageC#

Simple Neural Network for Flappy Bird

Overview

This project implements a simple neural network-based agent to play a Flappy Bird-inspired game environment. The goal is to train multiple agents to navigate through the environment and achieve the highest score possible by avoiding obstacles.

The project is implemented in C# using the Raylib library for rendering and visualization.

Features

  • Neural Network Agents: Each agent is equipped with a simple feedforward neural network that learns to play the game through reinforcement learning.
  • Replay Buffer: Agents use a replay buffer to store and learn from past experiences.
  • Training and Testing: The project includes both training and testing phases, allowing you to observe how well the agents have learned.
  • Multiple Agents: The environment supports multiple agents training simultaneously.
  • Customizable Parameters: You can adjust the number of agents, network structure, learning rate, and other hyperparameters.

Project Structure

  • Program.cs: The main entry point for the application. This file contains the logic for training and testing the agents.
  • FlappyBirdEnvironment.cs: Defines the game environment in which the agents operate. It includes the rendering logic, physics simulation, and interaction with agents.
  • Agent.cs: Defines the agent class, including the neural network, decision-making process, and training logic.
  • ReplayBuffer.cs: Implements the replay buffer that stores the agent's experiences for training.

Dependencies

  • Raylib-cs: A C# binding for Raylib, used for graphics and window management.
  • .NET Core 3.1 or later: The project is built on .NET Core, so you'll need a compatible version installed.

Installation

  1. Clone the repository:

    git clone https://github.com/yourusername/Simple_Neural_Network.git
cd Simple_Neural_Network

  2. Install dependencies:

    Make sure you have .NET Core and Raylib installed.

  3. Build and run the project:

    dotnet run

Usage

  • Training: The agents are trained for a specified number of episodes (episodes variable in Program.cs). During training, the agents learn to navigate through the game environment by avoiding obstacles.
  • Testing: After training, you can test the performance of the trained agents in the environment.
  • Rendering: The environment is rendered using Raylib, allowing you to visually observe the agents' performance.

Customization

You can customize various aspects of the project:

  • Network Structure: Modify the networkStructure array in Program.cs to change the architecture of the neural network.
  • Hyperparameters: Adjust the learning rate, discount factor, and replay buffer size in Program.cs.
  • Number of Agents: Change the numAgents variable to train multiple agents simultaneously.

Contributing

If you'd like to contribute to this project, feel free to fork the repository and submit a pull request with your changes.

Feel free to adjust the details or add any additional information specific to your project!