Struct Model 

pub struct Model {
    pub layers: Vec<Box<dyn Layer>>,
    pub optimizer: Optimizers,
}

A neural network model with a keras-inspired API

Fields

layers: Vec<Box<dyn Layer>>

optimizer: Optimizers

Implementations

impl Model

pub fn new() -> Model

Create a new model

Examples found in repository

examples/nn_full.rs (line 10)

fn main() {
    Logger::new().init().unwrap();
    let x = Tensor::linspace(-5.0, 5.0, 50);
    let func = &x * -5.0;
    let y = &func + &(Tensor::rand([50, 1]) * 0.4);

    let mut model = Model::new();
    model.add_layer(Linear::new(1, 10, Activations::None));
    model.add_layer(Linear::new(10, 10, Activations::None));
    model.add_layer(Linear::new(10, 1, Activations::None));

    model.compile(Optimizers::SGD);
    model.fit(&x, &y, 500, 0.00001, true);

    let x_pred = scalar!(1.0);
    let y_pred = model.predict(&x_pred);
    println!("Predict result (should be -5): {:?}", y_pred);
}

pub fn add_layer(&mut self, layer: Linear)

Add a layer to a model

Examples found in repository

examples/nn_full.rs (line 11)

fn main() {
    Logger::new().init().unwrap();
    let x = Tensor::linspace(-5.0, 5.0, 50);
    let func = &x * -5.0;
    let y = &func + &(Tensor::rand([50, 1]) * 0.4);

    let mut model = Model::new();
    model.add_layer(Linear::new(1, 10, Activations::None));
    model.add_layer(Linear::new(10, 10, Activations::None));
    model.add_layer(Linear::new(10, 1, Activations::None));

    model.compile(Optimizers::SGD);
    model.fit(&x, &y, 500, 0.00001, true);

    let x_pred = scalar!(1.0);
    let y_pred = model.predict(&x_pred);
    println!("Predict result (should be -5): {:?}", y_pred);
}

pub fn forward(&self, x: &Tensor) -> Tensor

Compute the forward pass of a model

pub fn parameters(&self) -> Vec<&Tensor>

Get the weights and biases of a model

pub fn compile(&mut self, optimizer: Optimizers)

Configure a model with an optimizer

Examples found in repository

examples/nn_full.rs (line 15)

fn main() {
    Logger::new().init().unwrap();
    let x = Tensor::linspace(-5.0, 5.0, 50);
    let func = &x * -5.0;
    let y = &func + &(Tensor::rand([50, 1]) * 0.4);

    let mut model = Model::new();
    model.add_layer(Linear::new(1, 10, Activations::None));
    model.add_layer(Linear::new(10, 10, Activations::None));
    model.add_layer(Linear::new(10, 1, Activations::None));

    model.compile(Optimizers::SGD);
    model.fit(&x, &y, 500, 0.00001, true);

    let x_pred = scalar!(1.0);
    let y_pred = model.predict(&x_pred);
    println!("Predict result (should be -5): {:?}", y_pred);
}

pub fn fit( &mut self, x: &Tensor, y: &Tensor, epochs: usize, lr: f64, debug: bool, )

Train a model

Examples found in repository

examples/nn_full.rs (line 16)

fn main() {
    Logger::new().init().unwrap();
    let x = Tensor::linspace(-5.0, 5.0, 50);
    let func = &x * -5.0;
    let y = &func + &(Tensor::rand([50, 1]) * 0.4);

    let mut model = Model::new();
    model.add_layer(Linear::new(1, 10, Activations::None));
    model.add_layer(Linear::new(10, 10, Activations::None));
    model.add_layer(Linear::new(10, 1, Activations::None));

    model.compile(Optimizers::SGD);
    model.fit(&x, &y, 500, 0.00001, true);

    let x_pred = scalar!(1.0);
    let y_pred = model.predict(&x_pred);
    println!("Predict result (should be -5): {:?}", y_pred);
}

pub fn predict(&self, x: &Tensor) -> Tensor

Make predictions from a model

Examples found in repository

examples/nn_full.rs (line 19)

fn main() {
    Logger::new().init().unwrap();
    let x = Tensor::linspace(-5.0, 5.0, 50);
    let func = &x * -5.0;
    let y = &func + &(Tensor::rand([50, 1]) * 0.4);

    let mut model = Model::new();
    model.add_layer(Linear::new(1, 10, Activations::None));
    model.add_layer(Linear::new(10, 10, Activations::None));
    model.add_layer(Linear::new(10, 1, Activations::None));

    model.compile(Optimizers::SGD);
    model.fit(&x, &y, 500, 0.00001, true);

    let x_pred = scalar!(1.0);
    let y_pred = model.predict(&x_pred);
    println!("Predict result (should be -5): {:?}", y_pred);
}