Struct rusty_machine::learning::gp::GaussianProcess

pub struct GaussianProcess<T: Kernel, U: MeanFunc> {
    pub noise: f64,
    /* private fields */
}

Gaussian Process struct

Gaussian process with generic kernel and deterministic mean function. Can be used for gaussian process regression with noise. Currently does not support classification.

Fields

noise: f64

The observation noise of the GP.

Implementations

impl<T: Kernel, U: MeanFunc> GaussianProcess<T, U>

pub fn new(ker: T, mean: U, noise: f64) -> GaussianProcess<T, U>

Construct a new Gaussian Process.

Examples

use rusty_machine::learning::gp;
use rusty_machine::learning::toolkit::kernel;

let ker = kernel::SquaredExp::default();
let mean = gp::ConstMean::default();
let gaussp = gp::GaussianProcess::new(ker, mean, 1e-3f64);

impl<T: Kernel, U: MeanFunc> GaussianProcess<T, U>

pub fn get_posterior(
    &self,
    inputs: &Matrix<f64>
) -> LearningResult<(Vector<f64>, Matrix<f64>)>

Compute the posterior distribution [UNSTABLE]

Requires the model to be trained first.

Outputs the posterior mean and covariance matrix.

Trait Implementations

impl<T: Debug + Kernel, U: Debug + MeanFunc> Debug for GaussianProcess<T, U>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for GaussianProcess<SquaredExp, ConstMean>

Construct a default Gaussian Process

The defaults are:

• Squared Exponential kernel.
• Zero-mean function.
• Zero noise.

Note that zero noise can often lead to numerical instability. A small value for the noise may be a better alternative.

fn default() -> GaussianProcess<SquaredExp, ConstMean>

Returns the “default value” for a type.

impl<T: Kernel, U: MeanFunc> SupModel<Matrix<f64>, Vector<f64>> for GaussianProcess<T, U>

fn predict(&self, inputs: &Matrix<f64>) -> LearningResult<Vector<f64>>

Predict output from inputs.

fn train(
    &mut self,
    inputs: &Matrix<f64>,
    targets: &Vector<f64>
) -> LearningResult<()>

Train the model using data and outputs.