About

Key Features

• Fully Connected Layers
• Convolution Layers
• Mini-Batch Gradient Descent
• Normalizations
• Model Saving/Loading to JSON

How To Use

use Sprouts::{Layer::{Layer, LayerType}, network::Network, activation::ActivationFunction::*, loss_function::LossType::*}

let layers = vec![
    Layer::dense([2, 3], Sigmoid),
    Layer::dense([3, 1], Sigmoid),
];

// Network::new(layers, learning_rate, batch_size, loss_function);
let nn = Network::new(layers, 0.2, 1, MSE);

//Prints network's loss and epoch progress in the terminal
nn.dense_train(true);

//data: Vec<[Inputs, Outputs]>
let data: Vec<[Vec<f64>; 2]> = vec![
    [vec![1.0, 0.0], vec![0.0]],
    [vec![0.0, 0.0], vec![1.0]],
    [vec![1.0, 1.0], vec![1.0]],
    [vec![0.0, 1.0], vec![0.0]],
];  

//dense_train(data, epochs)
nn.dense_train(data.clone(), 10000);

for i in 0..data.len() {
    println!("Input: {:?} || Output: {:?} || Target: {:?}",data[i][0].clone(), nn.dense_forward(data[i][0].clone()), data[i][1].clone());
}

As of now the only supported layers are conv and dense layers, pooling layers are next on the agenda.

will expound readme soon...

License

This project is licensed under the MIT License.