/**
 * Density-based clustering (DBSCAN) + k-nearest-neighbour classifier/regressor
 * (Phase 3 Task 4 — ML primitives).
 *
 * All three functions share brute-force Euclidean-distance neighbour search.
 * The library's exported `kdTree` (`typed/geometry.ts`) has no radius-query
 * method, so DBSCAN's ε-neighborhoods are computed O(n²) brute-force here —
 * correct, just not the asymptotically fastest option. A kd-tree range-search
 * (and a kd-tree-backed k-NN search) is future work if this becomes a hot path.
 */
/**
 * DBSCAN density-based clustering. A point is a **core point** if its
 * ε-neighborhood (including itself) has at least `minPts` members; clusters
 * are grown by expanding outward from core points through their neighbors
 * (density-reachability). Points reached only from a core point's
 * neighborhood but that are not themselves core are labeled as border points
 * of that cluster; points never reached are **noise** (`-1`).
 *
 * @param points - Row-vectors (n × d)
 * @param eps - Neighborhood radius (Euclidean)
 * @param minPts - Minimum neighborhood size (including the point itself) for a core point
 * @returns 0-based cluster label per point; `-1` marks noise
 *
 * @example
 * dbscan([[0, 0], [0.1, 0.1], [10, 10], [50, 50]], 1.0, 2)
 * // => [0, 0, -1, -1]  (one 2-point cluster, two noise points)
 */
export declare function dbscan(points: number[][], eps: number, minPts: number): number[];
/**
 * k-nearest-neighbour classifier: majority vote of the `k` closest training
 * points (Euclidean distance). Ties are broken by the label of the single
 * nearest point among the tied labels.
 *
 * @param train - Training row-vectors (n × d)
 * @param labels - Training labels (length n)
 * @param query - Query row-vectors (m × d)
 * @param k - Number of neighbors
 * @returns Predicted label per query row
 */
export declare function knnClassify(train: number[][], labels: (number | string)[], query: number[][], k: number): (number | string)[];
/**
 * k-nearest-neighbour regressor: mean of the `k` closest training targets
 * (Euclidean distance).
 *
 * @param train - Training row-vectors (n × d)
 * @param targets - Training targets (length n)
 * @param query - Query row-vectors (m × d)
 * @param k - Number of neighbors
 * @returns Predicted (mean) target per query row
 */
export declare function knnRegress(train: number[][], targets: number[], query: number[][], k: number): number[];
//# sourceMappingURL=dbscan-knn.d.ts.map