clustering-tfjs/dist/utils/affinity.d.ts

High-performance TypeScript clustering algorithms (K-Means, Spectral, Agglomerative) with TensorFlow.js acceleration and scikit-learn compatibility

import * as tf from '../tf-adapter';
/**
 * Computes the RBF (Gaussian) kernel affinity matrix for the given points.
 *
 *  A[i, j] = exp(-gamma * ||x_i - x_j||^2)
 *
 *  • The diagonal is guaranteed to be exactly 1 (because the distance is 0).
 *  • The result is symmetric by construction.
 *
 * The function is wrapped in `tf.tidy` so that all intermediate tensors are
 * automatically disposed of once the result tensor has been returned.
 */
export declare function compute_rbf_affinity(points: tf.Tensor2D, gamma?: number): tf.Tensor2D;
/**
 * Builds a (k-)nearest-neighbour adjacency / affinity matrix.
 *
 * For each sample the `k` closest neighbours are connected with affinity
 * value **1**. Self-loops are included to ensure connectivity, matching
 * sklearn's behavior. The final matrix is **symmetrised** via `max(A, Aᵀ)`
 * so that an edge is present when either sample appears in the other's
 * neighbourhood.
 *
 * The result is returned as a dense `tf.Tensor2D` containing zeros for
 * non-connected pairs.  While a sparse representation would be more memory
 * efficient, downstream TensorFlow.js ops (e.g. eigen-decomposition) currently
 * expect dense tensors.
 */
export declare function compute_knn_affinity(points: tf.Tensor2D, k: number, includeSelf?: boolean): tf.Tensor2D;
/**
 * Convenience wrapper that dispatches to the appropriate affinity builder
 * based on the provided `affinity` option.
 */
export declare function compute_affinity_matrix(points: tf.Tensor2D, options: {
    affinity: 'rbf';
    gamma?: number;
} | {
    affinity: 'nearest_neighbors';
    nNeighbors: number;
}): tf.Tensor2D;
//# sourceMappingURL=affinity.d.ts.map