import { ClassifierBase } from '../base';
import { Params } from '../base/estimator';
export interface QuadraticDiscriminantAnalysisProps {
    /** Class priors (in sorted-class order). Defaults to class frequencies. */
    priors?: number[];
    /** Regularization: per-class covariance eigenvalues become (1 - regParam) * s2 + regParam. */
    regParam?: number;
    /** Threshold on covariance eigenvalues for rank estimation. */
    tol?: number;
}
/**
 * Quadratic Discriminant Analysis matching sklearn's
 * `QuadraticDiscriminantAnalysis`: one Gaussian per class with its own
 * covariance, fitted through a per-class SVD of the centered class data
 * (never forming the covariance matrix), giving a quadratic decision
 * boundary log P(x|k) + log P(k).
 *
 * Deviation from sklearn: on rank-deficient class data with regParam = 0,
 * sklearn only warns about collinearity and then produces inf/NaN scores;
 * here directions whose covariance eigenvalue is <= tol are dropped from the
 * Mahalanobis distance and log-determinant, so results stay finite.
 */
export declare class QuadraticDiscriminantAnalysis extends ClassifierBase {
    private priors?;
    private regParam;
    private tol;
    private classes;
    private classPriors;
    private means;
    /** per class: kept covariance eigenvalues (descending) */
    private scalings;
    /** per class: kept principal directions, rotations[k][c] is a length-nFeatures vector */
    private rotations;
    private nFeatures;
    private fitted;
    constructor(props?: QuadraticDiscriminantAnalysisProps);
    getParams(): Params;
    fit(X: number[][], y: number[]): void;
    private assertFitted;
    /** Per-class log-posterior parts: -0.5 (Mahalanobis + logdet) + log prior. */
    private decisionScores;
    predict(testX: number[][]): number[];
    /** Softmax over the per-class log posteriors (sklearn parity). */
    predictProba(testX: number[][]): number[][];
    /**
     * Binary: 1-D signed scores for `classes[1]` (sklearn convention).
     * Multiclass: per-class log-posterior parts, shape [nSamples][nClasses].
     */
    decisionFunction(testX: number[][]): number[] | number[][];
    getClasses(): number[];
    getMeans(): number[][];
    getPriors(): number[];
    /** Per-class covariance matrices reconstructed from the kept eigenpairs. */
    getCovariance(): number[][][];
}