/**
 * Many numerical eigensolvers return eigenvectors in arbitrary order and with
 * an arbitrary global ± sign per vector.  For downstream algorithms (e.g.
 * Spectral Clustering) we require a stable ordering and sign convention so
 * that identical input always produces identical embeddings.
 *
 * The convention implemented here matches scikit-learn:
 *   1. Eigen-pairs are sorted by ascending eigen-value.
 *   2. For every eigen-vector the component with the largest absolute value
 *      is made positive by optionally multiplying the vector by −1.
 *
 * The function operates purely on native JavaScript arrays to avoid pulling
 * TensorFlow.js into low-level utilities and reduce garbage generation.
 */
export interface EigenPairInput {
    eigenvalues: number[];
    eigenvectors: number[][];
}
export interface EigenPairOutput {
    eigenvalues: number[];
    eigenvectors: number[][];
}
export declare function deterministic_eigenpair_processing(input: EigenPairInput): EigenPairOutput;