Struct rusty_machine::learning::optim::grad_desc::StochasticGD

pub struct StochasticGD {
    pub alpha: f64,
    pub mu: f64,
    pub iters: usize,
}

Stochastic Gradient Descent algorithm.

Uses basic momentum to control the learning rate.

Fields

alpha: f64

Controls the momentum of the descent

mu: f64

The square root of the raw learning rate.

iters: usize

The number of passes through the data.

Methods

impl StochasticGD

fn new(alpha: f64, mu: f64, iters: usize) -> StochasticGD

Construct a stochastic gradient descent algorithm.

Requires the learning rate, momentum rate and iteration count to be specified.

Examples

use rusty_machine::learning::optim::grad_desc::StochasticGD;

let sgd = StochasticGD::new(0.1, 0.3, 5);

Trait Implementations

impl Default for StochasticGD

fn default() -> StochasticGD

Constructs a stochastic gradient descent algorithm with default settings.

Uses 5 iterations, momentum of 0.1 and rate of 0.3.

impl<M: Optimizable<Data=Matrix<f64>, Target=Matrix<f64>>> OptimAlgorithm<M> for StochasticGD

fn optimize(&self, model: &M, start: &[f64], data: &M::Data, outputs: &M::Target) -> Vec<f64>

Return the optimized parameter using gradient optimization.