/**
 * Stacked generalization meta-estimators, mirroring sklearn's
 * `StackingClassifier` and `StackingRegressor`.
 *
 * The final estimator is trained on OUT-OF-FOLD predictions of the base
 * estimators: the data is split with (Stratified)KFold, each base estimator
 * is `clone()`d per fold, fit on the fold's training part, and its
 * predictions on the held-out part fill that fold's rows of the meta-feature
 * matrix. The base estimators themselves are then refit on the full data and
 * used to build meta-features at inference time.
 */
import { ClassifierBase, RegressorBase } from '../base';
import { BaseEstimator, Params } from '../base/estimator';
import { ClassifierLike, RegressorLike, NamedEstimator } from './voting';
export type StackMethod = 'auto' | 'predictProba' | 'decisionFunction' | 'predict';
export interface StackingClassifierProps {
    /** Named base estimators; addressable as `name__param` in setParams. */
    estimators: NamedEstimator<ClassifierLike>[];
    /**
     * Meta-estimator fit on the out-of-fold meta-features. Defaults to
     * `LogisticRegression()` (binary-only in this library — pass a
     * multiclass-capable classifier for >2 classes). Addressable as
     * `finalEstimator__param`.
     */
    finalEstimator?: ClassifierLike;
    /** StratifiedKFold split count for the out-of-fold predictions. */
    cv?: number;
    /** How base estimators produce meta-features (see resolveStackMethod). */
    stackMethod?: StackMethod;
    /** Append the original features after the meta-features. */
    passthrough?: boolean;
}
export declare class StackingClassifier extends ClassifierBase {
    private estimators;
    private finalEstimator;
    private cv;
    private stackMethod;
    private passthrough;
    private classes;
    private resolvedMethods;
    private fitted;
    constructor(props: StackingClassifierProps);
    getParams(): Params;
    /** Supports own params plus `name__param` / `finalEstimator__param`. */
    setParams(params: Params): this;
    getEstimator(name: string): BaseEstimator;
    get namedEstimators(): Record<string, BaseEstimator>;
    getClasses(): number[];
    fit(X: number[][], y: number[], sampleWeight?: number[]): void;
    /** Meta-features for X from the full-data base estimators. */
    transform(X: number[][]): number[][];
    predict(X: number[][]): number[];
    predictProba(X: number[][]): number[][];
    decisionFunction(X: number[][]): number[] | number[][];
    private assertFitted;
}
export interface StackingRegressorProps {
    /** Named base estimators; addressable as `name__param` in setParams. */
    estimators: NamedEstimator<RegressorLike>[];
    /**
     * Meta-estimator fit on the out-of-fold predictions. Defaults to
     * `RidgeRegression()`. Addressable as `finalEstimator__param`.
     */
    finalEstimator?: RegressorLike;
    /** KFold split count for the out-of-fold predictions. */
    cv?: number;
    /** Append the original features after the meta-features. */
    passthrough?: boolean;
}
export declare class StackingRegressor extends RegressorBase {
    private estimators;
    private finalEstimator;
    private cv;
    private passthrough;
    private fitted;
    constructor(props: StackingRegressorProps);
    getParams(): Params;
    /** Supports own params plus `name__param` / `finalEstimator__param`. */
    setParams(params: Params): this;
    getEstimator(name: string): BaseEstimator;
    get namedEstimators(): Record<string, BaseEstimator>;
    fit(X: number[][], y: number[], sampleWeight?: number[]): void;
    /** Meta-features for X from the full-data base estimators. */
    transform(X: number[][]): number[][];
    predict(X: number[][]): number[];
}