mario-so/neural_rs

Neural network from scratch that learns to do XOR gate.

🧠 Simple Neural Network in Rust 🤖

Welcome to the neural_rs project implemented in Rust! 🦀

This program demonstrates the fundamentals of a neural network with forward and backpropagation from scratch.

The code is inspired from this video in which the guy coded the same but in C language.

🚀 Getting Started

Prerequisites

Ensure you have Rust installed. You can install Rust using rustup:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

Installation

1. Clone the repository:

   git clone https://github.com/Mario-SO/neural_rs/tree/main
   cd neural_rs

2. Run the project:

   cargo run

🎯 Purpose

This simple neural network aims to classify XOR gate outputs 🌓. The network consists of:

• An input layer with 2 neurons
• One hidden layer with 2 neurons
• An output layer with 1 neuron

📝 How It Works

Initialization

• Weights and Biases: Initialized randomly between 0 and 1.

Forward Pass

1. Hidden Layer: Calculates activations using input data, weights, and biases, then applies the sigmoid function.
2. Output Layer: Computes the final output using hidden layer activations, weights, and biases, then applies the sigmoid function.

Backpropagation

• Adjusts weights and biases to minimize the error between the predicted and actual outputs using the learning rate and derivatives of the sigmoid function.

📊 Training & Accuracy

• Trains the network over several epochs and prints the accuracy on the training set after training.
• Displays final weights and biases after training is complete.

📈 Example Output

Each training iteration shows:

• Input Data
• Predicted Output
• Expected Output

After training, it prints the final accuracy, weights, and biases.

Example:

Input: [0.0, 0.0] -> Predicted output: [0.045] (Expected: [0.0])
Input: [0.0, 1.0] -> Predicted output: [0.912] (Expected: [1.0])
...
Final accuracy: 100.00%
Final hidden weights: [[0.2, 0.3], [0.4, 0.1]]
Final output weights: [[0.5, 0.7]]
...

🛠️ Key Functions

1. initialize_weights: Initializes weights randomly.
2. sigmoid: Sigmoid activation function.
3. sigmoid_derivative: Derivative of the sigmoid function for backpropagation.

🤝 Contributions

Feel free to fork this repository and contribute by submitting a pull request. Please ensure all changes are well tested.

---

Happy coding! 🎉

🌟 Star this repository if you found it helpful!