import * as tf from '../backend/adapter';
import { SparseMatrix } from './sparse';
/**
 *  A[i, j] = exp(-gamma * ||x_i - x_j||^2)
 */
export declare function compute_rbf_affinity(points: tf.Tensor2D, gamma?: number): tf.Tensor2D;
/**
 * Self-loops are included by default to ensure connectivity (sklearn behaviour).
 * Symmetrised via `0.5 * (A + Aᵀ)`: mutual edges → 1.0, asymmetric → 0.5.
 *
 * Returns a dense tensor because downstream ops (eigen-decomposition) expect
 * dense input; use `compute_sparse_knn_affinity` for the CSR form.
 */
export declare function compute_knn_affinity(points: tf.Tensor2D, k: number, include_self?: boolean): tf.Tensor2D;
/**
 * Symmetrised as `0.5 * (A + Aᵀ)`, matching sklearn's SpectralClustering
 * connectivity path.
 */
export declare function compute_sparse_knn_affinity(points: tf.Tensor2D, k: number, include_self?: boolean): SparseMatrix;
/**
 *   A[i, j] = 1 - cosine_distance(x_i, x_j)
 *
 * Natural similarity for direction-dominated, magnitude-noisy data (text
 * embeddings, TF-IDF vectors). Diagonal is forced to exactly 1.
 */
export declare function compute_cosine_affinity(points: tf.Tensor2D): tf.Tensor2D;