export declare const EPSILON = 1e-10;
export declare function clampProbability(p: number): number;
export declare function clampProbability(p: number[]): number[];
export declare function sigmoid(x: number): number;
export declare function sigmoid(x: number[]): number[];
export declare function logit(p: number): number;
export declare function logit(p: number[]): number[];
export type TrainingMode = "balanced" | "prior_aware" | "prior_free";
export interface FitOptions {
    learningRate?: number;
    maxIterations?: number;
    tolerance?: number;
    mode?: TrainingMode;
    tfs?: number[];
    docLenRatios?: number[];
}
export interface UpdateOptions {
    learningRate?: number;
    momentum?: number;
    decayTau?: number;
    maxGradNorm?: number;
    avgDecay?: number;
    mode?: TrainingMode;
    tf?: number | number[];
    docLenRatio?: number | number[];
}
export declare class BayesianProbabilityTransform {
    alpha: number;
    beta: number;
    readonly baseRate: number | null;
    private _logitBaseRate;
    protected _trainingMode: TrainingMode;
    protected _nUpdates: number;
    protected _gradAlphaEMA: number;
    protected _gradBetaEMA: number;
    protected _alphaAvg: number;
    protected _betaAvg: number;
    protected _priorFn: ((score: number | number[], tf: number | number[], docLenRatio: number | number[]) => number | number[]) | null;
    constructor(alpha?: number, beta?: number, baseRate?: number | null, priorFn?: ((score: number | number[], tf: number | number[], docLenRatio: number | number[]) => number | number[]) | null);
    get averagedAlpha(): number;
    get averagedBeta(): number;
    get nUpdates(): number;
    get gradAlphaEMA(): number;
    likelihood(score: number): number;
    likelihood(score: number[]): number[];
    static tfPrior(tf: number): number;
    static tfPrior(tf: number[]): number[];
    static normPrior(docLenRatio: number): number;
    static normPrior(docLenRatio: number[]): number[];
    static compositePrior(tf: number, docLenRatio: number): number;
    static compositePrior(tf: number[], docLenRatio: number[]): number[];
    static posterior(likelihoodVal: number, prior: number, baseRate?: number | null): number;
    static posterior(likelihoodVal: number[], prior: number[], baseRate?: number | null): number[];
    scoreToProbability(score: number, tf: number, docLenRatio: number): number;
    scoreToProbability(score: number[], tf: number[], docLenRatio: number[]): number[];
    wandUpperBound(bm25UpperBound: number): number;
    wandUpperBound(bm25UpperBound: number[]): number[];
    fit(scores: number[], labels: number[], options?: FitOptions): void;
    update(score: number | number[], label: number | number[], options?: UpdateOptions): void;
}
export interface TemporalFitOptions extends FitOptions {
    timestamps?: number[];
}
export declare class TemporalBayesianTransform extends BayesianProbabilityTransform {
    private _decayHalfLife;
    private _decayRate;
    private _timestamp;
    constructor(alpha?: number, beta?: number, baseRate?: number | null, decayHalfLife?: number);
    get decayHalfLife(): number;
    get timestamp(): number;
    fit(scores: number[], labels: number[], options?: TemporalFitOptions): void;
    update(score: number | number[], label: number | number[], options?: UpdateOptions): void;
}
//# sourceMappingURL=probability.d.ts.map