Struct slow_nn::Network

pub struct Network { /* fields omitted */ }

Examples

let mut net = network::Network::dense(bias, num_inputs, num_outputs, &num_hiddens);
let pred = net.predict(&input, activation);

Methods

impl Network

pub fn from_conns(
    bias: f64,
    inputs: usize,
    outputs: usize,
    hidden: usize,
    conns: &[Connection]
) -> Self

Creates a network from a slice of Connections

pub fn dense(bias: f64, inputs: usize, outputs: usize, layers: &[usize]) -> Self

Creates a fully connected network with random weights

pub fn predict(&self, input: &[f64], activation: fn(_: f64) -> f64) -> Vec<f64>

Does a forward pass

pub fn train(
    &mut self,
    input: &[f64],
    expected: &[f64],
    activation: fn(_: f64) -> f64,
    deactivation: fn(_: f64) -> f64,
    loss: fn(_: f64, _: f64) -> f64,
    dloss: fn(_: f64, _: f64) -> f64,
    lr: f64
) -> f64

Does both forward and backward pass and updates the weights