import { BaseEstimator, Params } from '../base/estimator';
/**
 * Structural contract for the wrapped classifier. Validated at runtime by
 * shape so any estimator following the BaseEstimator contract works,
 * including binary-only classifiers (that is the point of these wrappers).
 */
export interface ClassifierLike extends BaseEstimator {
    fit(X: number[][], y: number[], sampleWeight?: number[]): void;
    predict(X: number[][]): number[];
    predictProba?(X: number[][]): number[][];
    decisionFunction?(X: number[][]): number[] | number[][];
}
export declare function validateClassifierEstimator(estimator: unknown, wrapper: string): ClassifierLike;
export declare function validateFitInputs(X: number[][], y: number[], wrapper: string): void;
export declare function sortedUniqueLabels(y: number[]): number[];
/**
 * Split params into the wrapper's own params and nested params addressed as
 * `estimator__<param>` (grid-search style, mirroring Pipeline's
 * `step__param` addressing).
 */
export declare function splitEstimatorParams(params: Params, className: string): {
    own: Params;
    nested: Params;
};
/** Flatten a member's decisionFunction output to one score per sample. */
export declare function flattenDecision(d: number[] | number[][]): number[];
/**
 * Per-sample score of the positive class from a fitted binary member trained
 * on labels {0, 1}: decisionFunction (flattened) when available, else
 * predictProba's positive-class column, else the raw predict output (0/1
 * votes). decisionFunction comes first to mirror sklearn's _predict_binary
 * and to stay consistent with OneVsOneClassifier's confidence source.
 */
export declare function binaryScores(est: ClassifierLike, X: number[][]): number[];
/** Index of the row maximum; the first maximum wins on ties. */
export declare function argmax(row: number[]): number;