Struct juggernaut::nn::NeuralNetwork

pub struct NeuralNetwork { /* fields omitted */ }

Represents a Neural Network with layers, inputs and outputs

Methods

impl NeuralNetwork

fn new() -> NeuralNetwork

fn set_shuffle_data(&mut self, enable: bool)

To set shuffle data flag Enabling this option shuffles data before each iteration

fn set_cost_function<T>(&mut self, cost_function: T) where
    T: 'static + CostFunction, 

To set a cost function for the network

fn on_error<FN>(&mut self, callback_fn: FN) where
    FN: 'static + Fn(f64), 

To add a callback function and receive the errors of the network during training process Please note that there is another function that basically calcualtes the error value

fn on_epoch<FN>(&mut self, callback_fn: FN) where
    FN: 'static + Fn(&NeuralNetwork), 

To add a callback function to get called after each epoch

fn add_layer(&mut self, layer: NeuralLayer)

To add a new layer to the network

Example:

use juggernaut::sample::Sample;
use juggernaut::nl::NeuralLayer;
use juggernaut::nn::NeuralNetwork;
use juggernaut::activation::Activation;
use juggernaut::activation::Sigmoid;

let mut test = NeuralNetwork::new();

// 1st layer = 4 neurons - 2 inputs
let nl1 = NeuralLayer::new(4, 2, Sigmoid::new());

test.add_layer(nl1);

fn get_layers(&self) -> &Vec<NeuralLayer>

To get the layers of the network

fn forward(&self, sample: &Sample) -> Vec<Matrix>

This is the forward method of the network which calculates the random weights and multiplies the inputs of given samples to the weights matrix. Thinks.

fn evaluate(&self, sample: &Sample) -> Matrix

Use this function to evaluate a trained neural network

This function simply passes the given sample to the forward function and returns the output of last layer

fn train(&mut self, samples: Vec<Sample>, epochs: i32, learning_rate: f64)

To train the network. It calls the forward pass and updates the weights using backpropagation