import { RegressorBase } from '../base';
import { Params } from '../base/estimator';
import { RegressionLossName, SGDLearningRate, SGDPenalty } from './sgdBase';
export interface SGDRegressorProps {
    loss?: RegressionLossName;
    /** Threshold of the huber / (squared) epsilon-insensitive losses. */
    epsilon?: number;
    penalty?: SGDPenalty;
    alpha?: number;
    l1Ratio?: number;
    fitIntercept?: boolean;
    maxIter?: number;
    tol?: number | null;
    shuffle?: boolean;
    randomState?: number;
    learningRate?: SGDLearningRate;
    eta0?: number;
    powerT?: number;
    nIterNoChange?: number;
}
/**
 * Linear regressor trained with plain (non-averaged) stochastic gradient
 * descent, mirroring scikit-learn's `SGDRegressor` (default schedule
 * `invscaling` with eta0=0.01, powerT=0.25).
 */
export declare class SGDRegressor extends RegressorBase {
    private loss;
    private epsilon;
    private penalty;
    private alpha;
    private l1Ratio;
    private fitIntercept;
    private maxIter;
    private tol;
    private shuffle;
    private randomState?;
    private learningRate;
    private eta0;
    private powerT;
    private nIterNoChange;
    private coefState;
    private intercept;
    private fitted;
    private nIter;
    constructor(props?: SGDRegressorProps);
    getParams(): Params;
    fit(trainX: number[][], trainY: number[]): void;
    predict(testX: number[][]): number[];
    getCoef(): number[];
    get coef(): number[];
    getIntercept(): number;
    /** Number of epochs run in the last fit. */
    getNIter(): number;
}