Meuron

A modular rust written library for training simple Neuronal Networks.

Features

• Modular layer system
• Multiple activation functions (ReLU, Sigmoid, Softmax)
• Multiple cost functions (MSE, CrossEntropy, BinaryCrossEntropy)
• Easy to extend with custom layers and activations

Quick Start

Add to your Cargo.toml:

[dependencies]
meuron = "0.1"

Basic Example

use meuron::{NeuralNetwork, layer::DenseLayer, activation::Sigmoid, cost::MSE};
use ndarray::Array2;

fn main() {
    // Create a simple 2-layer network
    let layer1 = DenseLayer::new(784, 128, Sigmoid);
    let layer2 = DenseLayer::new(128, 10, Sigmoid);

    let mut nn = NeuralNetwork::new(
        vec![layer1, layer2],
        MSE,
    );

    // Train the network
    nn.train(&train_data, &train_labels, 0.01, 10, 32);

    // Save the model
    nn.save("model.bin").unwrap();

    // Load later
    let loaded_nn = NeuralNetwork::load("model.bin", MSE).unwrap();
}

Available Components

Activations

• ReLU
• Sigmoid
• Softmax
• Tanh

Cost Functions

• MSE
• CrossEntropy
• BinaryCrossEntropy

Layers

• DenseLayer

Examples

See the examples/ directory:

cargo run --example mnist --release

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.