import { ClassifierBase } from '../base';
import { Params } from '../base/estimator';
export interface LinearSVCProps {
    C?: number;
    maxIter?: number;
    /** @deprecated ignored — Pegasos uses the schedule eta_t = 1/(lambda*t) */
    learningRate?: number;
    tol?: number;
    randomState?: number;
}
/**
 * Linear SVM trained with Pegasos (primal stochastic subgradient descent).
 *
 * Objective per class (one-vs-rest): lambda/2 ||w||^2 + (1/n) sum hinge,
 * with lambda = 1/(n*C) so the C semantics match sklearn's LinearSVC.
 * Step size follows the Pegasos schedule eta_t = 1/(lambda*t); the bias is
 * not regularized. maxIter counts epochs; training stops early when the
 * objective's relative improvement falls below tol.
 */
export declare class LinearSVC extends ClassifierBase {
    private C;
    private maxIter;
    /** @deprecated kept only so params round-trip; the Pegasos schedule ignores it */
    private learningRate?;
    private tol;
    private randomState?;
    private classes;
    private weights;
    private biases;
    constructor(props?: LinearSVCProps);
    getParams(): Params;
    private objective;
    private trainBinary;
    fit(trainX: number[][], trainY: number[]): void;
    predict(testX: number[][]): number[];
    get coef(): number[][];
}