export declare function cosineToProbability(score: number): number;
export declare function cosineToProbability(score: number[]): number[];
export declare function probNot(prob: number): number;
export declare function probNot(prob: number[]): number[];
export declare function probAnd(probs: number[]): number;
export declare function probAnd(probs: number[][]): number[];
export declare function probOr(probs: number[]): number;
export declare function probOr(probs: number[][]): number[];
export declare function resolveAlpha(alpha: number | "auto" | undefined, defaultValue: number): number;
export declare function logOddsConjunction(probs: number[], alpha?: number | "auto", weights?: number[], gating?: string, gatingBeta?: number): number;
export declare function logOddsConjunction(probs: number[][], alpha?: number | "auto", weights?: number[], gating?: string, gatingBeta?: number): number[];
export declare function balancedLogOddsFusion(sparseProbs: number[], denseSimilarities: number[], weight?: number): number[];
export declare class LearnableLogOddsWeights {
    private _nSignals;
    private _alpha;
    private _baseRate;
    private _logitBaseRate;
    private _logits;
    private _nUpdates;
    private _gradLogitsEMA;
    private _weightsAvg;
    constructor(nSignals: number, alpha?: number | "auto", baseRate?: number | null);
    get nSignals(): number;
    get alpha(): number;
    get baseRate(): number | null;
    get weights(): number[];
    get averagedWeights(): number[];
    combine(probs: number[], useAveraged?: boolean): number;
    combine(probs: number[][], useAveraged?: boolean): number[];
    fit(probs: number[][], labels: number[], options?: {
        learningRate?: number;
        maxIterations?: number;
        tolerance?: number;
    }): void;
    update(probs: number[] | number[][], label: number | number[], options?: {
        learningRate?: number;
        momentum?: number;
        decayTau?: number;
        maxGradNorm?: number;
        avgDecay?: number;
    }): void;
}
export declare class AttentionLogOddsWeights {
    private _nSignals;
    private _nQueryFeatures;
    private _alpha;
    private _normalize;
    private _baseRate;
    private _logitBaseRate;
    private _W;
    private _b;
    private _nUpdates;
    private _gradWEMA;
    private _gradBEMA;
    private _WAvg;
    private _bAvg;
    constructor(nSignals: number, nQueryFeatures: number, alpha?: number | "auto", normalize?: boolean, seed?: number, baseRate?: number | null);
    get nSignals(): number;
    get nQueryFeatures(): number;
    get alpha(): number;
    get baseRate(): number | null;
    get normalize(): boolean;
    private static _normalizeLogits;
    get weightsMatrix(): number[][];
    private _computeWeights;
    combine(probs: number[], queryFeatures: number[], useAveraged?: boolean): number;
    combine(probs: number[][], queryFeatures: number[][], useAveraged?: boolean): number[];
    fit(probs: number[][], labels: number[], queryFeatures: number[][], options?: {
        queryIds?: number[];
        learningRate?: number;
        maxIterations?: number;
        tolerance?: number;
    }): void;
    update(probs: number[] | number[][], label: number | number[], queryFeatures: number[] | number[][], options?: {
        learningRate?: number;
        momentum?: number;
        decayTau?: number;
        maxGradNorm?: number;
        avgDecay?: number;
    }): void;
    computeUpperBounds(upperBoundProbs: number[][], queryFeatures: number[] | number[][], useAveraged?: boolean): number[];
    prune(probs: number[][], queryFeatures: number[] | number[][], threshold: number, upperBoundProbs?: number[][], useAveraged?: boolean): {
        survivingIndices: number[];
        fusedProbabilities: number[];
    };
}
export declare class MultiHeadAttentionLogOddsWeights {
    private _nHeads;
    private _heads;
    constructor(nHeads: number, nSignals: number, nQueryFeatures: number, alpha?: number | "auto", normalize?: boolean);
    get nHeads(): number;
    get heads(): AttentionLogOddsWeights[];
    combine(probs: number[], queryFeatures: number[], useAveraged?: boolean): number;
    combine(probs: number[][], queryFeatures: number[] | number[][], useAveraged?: boolean): number[];
    fit(probs: number[][], labels: number[], queryFeatures: number[][], options?: {
        queryIds?: number[];
        learningRate?: number;
        maxIterations?: number;
        tolerance?: number;
    }): void;
    update(probs: number[] | number[][], label: number | number[], queryFeatures: number[] | number[][], options?: {
        learningRate?: number;
        momentum?: number;
        decayTau?: number;
        maxGradNorm?: number;
        avgDecay?: number;
    }): void;
    computeUpperBounds(upperBoundProbs: number[][], queryFeatures: number[] | number[][], useAveraged?: boolean): number[];
    prune(probs: number[][], queryFeatures: number[] | number[][], threshold: number, upperBoundProbs?: number[][], useAveraged?: boolean): {
        survivingIndices: number[];
        fusedProbabilities: number[];
    };
}
//# sourceMappingURL=fusion.d.ts.map