/** Supported covariance kernels. */
export type GPKernel = 'rbf' | 'matern32' | 'matern52';
/** Options for {@link gaussianProcessRegression}. */
export interface GPOptions {
    /** Covariance kernel. Default `'rbf'` (squared-exponential). */
    kernel?: GPKernel;
    /** Kernel length-scale ℓ (> 0). Default `1`. */
    lengthScale?: number;
    /** Signal variance σ_f² (kernel amplitude, > 0). Default `1`. */
    signalVariance?: number;
    /** I.i.d. Gaussian noise variance α added to the diagonal (≥ 0). Default `1e-10`. */
    noise?: number;
}
/** Posterior prediction at a set of test points. */
export interface GPPrediction {
    /** Posterior mean at each test point. */
    mean: number[];
    /** Posterior variance at each test point (≥ 0). */
    variance: number[];
    /** Posterior standard deviation at each test point (= √variance). */
    std: number[];
}
/** A fitted Gaussian-process regressor. */
export interface GPModel {
    /** Predict the posterior mean/variance/std at the given test points. */
    predict(Xstar: readonly (readonly number[])[]): GPPrediction;
    /** Log marginal likelihood log p(y | X) of the training data under the prior. */
    logMarginalLikelihood: number;
    /** The resolved kernel name. */
    kernel: GPKernel;
}
/**
 * Fit a Gaussian-process regressor to training points `X` (array of
 * length-d feature vectors) and targets `y`, then return a model exposing
 * `.predict(Xstar)` (posterior mean/variance/std) and the log marginal
 * likelihood.
 *
 * @example
 * const gp = gaussianProcessRegression(
 *   [[-4], [-3], [-1], [0], [2]],
 *   [-2, 0, 1, 2, -1],
 *   { kernel: 'rbf', lengthScale: 1.2, signalVariance: 1.5, noise: 1e-2 }
 * );
 * gp.predict([[-0.5], [10]]); // mean ≈ [1.5872, ~0], std ≈ [0.1453, 1.2247]
 */
export declare function gaussianProcessRegression(X: readonly (readonly number[])[], y: readonly number[], options?: GPOptions): GPModel;
/** Alias of {@link gaussianProcessRegression}. */
export declare const gpRegression: typeof gaussianProcessRegression;
//# sourceMappingURL=gaussian-process.d.ts.map