Struct ConvolutionProcess 

pub struct ConvolutionProcess<S = Untrained> { /* private fields */ }

Convolution Process (a.k.a. Convolved / Dependent Gaussian Process) for multi-output regression.

See the module documentation for the generative model and the closed-form induced covariance this type trains against. Use the builder methods below to configure it, then call Fit::fit with one input matrix and one target vector per output.

Examples

use sklears_gaussian_process::ConvolutionProcess;
use sklears_core::traits::{Fit, Predict};
use scirs2_core::ndarray::array;

let x1 = array![[0.0], [1.0], [2.0], [3.0]];
let y1 = array![0.0, 1.0, 4.0, 9.0];
let x2 = array![[0.5], [2.5]];
let y2 = array![2.0, 17.0];

let cp = ConvolutionProcess::new().n_latent(1).noise_variance(1e-6);
let fitted = cp
    .fit(&vec![x1, x2], &vec![y1, y2])
    .expect("fit should succeed with valid training data");
let predictions = fitted
    .predict(&vec![array![[1.5]], array![[1.0]]])
    .expect("predict should succeed on a trained model");
assert_eq!(predictions.len(), 2);

Implementations

impl ConvolutionProcess<Untrained>

pub fn new() -> Self

Create a new, unconfigured Convolution Process (n_latent = 1 by default).

pub fn n_latent(self, n_latent: usize) -> Self

Set the number of shared latent processes Q.

pub fn default_length_scale(self, length_scale: f64) -> Self

Set the default per-dimension length scale used when Self::length_scales is not supplied explicitly.

pub fn noise_variance(self, noise_variance: f64) -> Self

Set the default (uniform, per-output) noise variance used when Self::noise_variances is not supplied explicitly.

pub fn length_scales(self, length_scales: Vec<Vec<Array1<f64>>>) -> Self

Explicitly set the [output][latent] -> per-dimension length scale table. The outer Vec must have one entry per output (matching the length of the X passed to Fit::fit), each inner Vec one entry per latent process (matching Self::n_latent), and each Array1 one length scale per input dimension. Validated (and defaulted, if omitted) at fit time.

pub fn output_variances(self, output_variances: Vec<Vec<f64>>) -> Self

Explicitly set the [output][latent] -> signal variance table (same outer/ inner shape as Self::length_scales, minus the per-dimension axis).

pub fn noise_variances(self, noise_variances: Vec<f64>) -> Self

Explicitly set a per-output noise variance (one entry per output).

impl ConvolutionProcess<ConvolutionProcessTrained>

pub fn trained_state(&self) -> &ConvolutionProcessTrained

Access the full trained state.

pub fn n_outputs(&self) -> usize

Number of outputs the model was fitted on.

pub fn n_latent(&self) -> usize

Number of shared latent processes Q.

pub fn log_marginal_likelihood(&self) -> f64

Log marginal likelihood of the training data under the fitted model.

pub fn noise_variances(&self) -> &[f64]

Resolved per-output noise variance actually used for training (one entry per output, in fit order).

pub fn induced_covariance( &self, output_d: usize, output_dp: usize, x: &ArrayView1<'_, f64>, xp: &ArrayView1<'_, f64>, ) -> SklResult<f64>

Evaluate the model’s induced covariance k_{d,d'}(x, x') between output output_d and output output_dp (which may be equal) at arbitrary points, using the fitted (or defaulted) hyperparameters. See the module documentation for the closed-form formula being evaluated.

This is the direct, low-level way to inspect whether – and how strongly – the model represents correlation between two outputs: for two genuinely independent single-output GPs this would be identically zero everywhere, whereas a convolution process sharing at least one latent process between output_d and output_dp gives a nontrivial value.

pub fn training_covariance_matrix(&self) -> Array2<f64>

Rebuild the full (noise-free) stacked multi-output covariance matrix over the training points, using the fitted hyperparameters. This is the matrix whose positive semi-definiteness is the key theoretical guarantee of the convolution-process construction (see module docs); exposed for diagnostics and testing.

pub fn predict_with_std( &self, X_test: &[Array2<f64>], ) -> SklResult<Vec<(Array1<f64>, Array1<f64>)>>

Predict the posterior mean and standard deviation for each output at the given per-output test points. X_test must have exactly n_outputs() entries (one design matrix per output, in the same order as at fit time); pass a zero-row matrix for any output you are not interested in predicting.

Trait Implementations

impl<S: Clone> Clone for ConvolutionProcess<S>

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<S: Debug> Debug for ConvolutionProcess<S>

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

Formats the value using the given formatter.

impl Default for ConvolutionProcess<Untrained>

fn default() -> Self

Returns the “default value” for a type.

impl Estimator for ConvolutionProcess<Untrained>

type Config = ConvolutionProcessConfig

Configuration type for the estimator

type Error = SklearsError

Error type for the estimator

type Float = f64

The numeric type used by this estimator

fn config(&self) -> &Self::Config

Get estimator configuration

fn validate_config(&self) -> Result<(), SklearsError>

Validate estimator configuration with detailed error context

fn check_compatibility( &self, n_samples: usize, n_features: usize, ) -> Result<(), SklearsError>

Check if estimator is compatible with given data dimensions

fn metadata(&self) -> EstimatorMetadata

Get estimator metadata

impl Estimator for ConvolutionProcess<ConvolutionProcessTrained>

type Config = ConvolutionProcessConfig

Configuration type for the estimator

type Error = SklearsError

Error type for the estimator

type Float = f64

The numeric type used by this estimator

fn config(&self) -> &Self::Config

Get estimator configuration

fn validate_config(&self) -> Result<(), SklearsError>

Validate estimator configuration with detailed error context

fn check_compatibility( &self, n_samples: usize, n_features: usize, ) -> Result<(), SklearsError>

Check if estimator is compatible with given data dimensions

fn metadata(&self) -> EstimatorMetadata

Get estimator metadata

impl Fit<Vec<ArrayBase<OwnedRepr<f64>, Dim<[usize; 2]>>>, Vec<ArrayBase<OwnedRepr<f64>, Dim<[usize; 1]>>>> for ConvolutionProcess<Untrained>

type Fitted = ConvolutionProcess<ConvolutionProcessTrained>

The fitted model type

fn fit( self, X: &Vec<Array2<f64>>, Y: &Vec<Array1<f64>>, ) -> SklResult<Self::Fitted>

Fit the model to the provided data with validation

fn fit_with_validation( self, x: &X, y: &Y, _x_val: Option<&X>, _y_val: Option<&Y>, ) -> Result<(Self::Fitted, FitMetrics), SklearsError>

where Self: Sized,

Fit with custom validation and early stopping

impl Predict<Vec<ArrayBase<OwnedRepr<f64>, Dim<[usize; 2]>>>, Vec<ArrayBase<OwnedRepr<f64>, Dim<[usize; 1]>>>> for ConvolutionProcess<ConvolutionProcessTrained>

fn predict(&self, X: &Vec<Array2<f64>>) -> SklResult<Vec<Array1<f64>>>

Make predictions on the provided data

fn predict_with_uncertainty( &self, x: &X, ) -> Result<(Output, UncertaintyMeasure), SklearsError>

Make predictions with confidence intervals