@thi.ng/text-analysis/cluster.d.ts

Text tokenization, transformation & analysis transducers, utilities, stop words, porter stemming, vector encodings, similarities

import { Untransformed } from "@thi.ng/distance/untransformed";
import { type KMeansOpts } from "@thi.ng/k-means";
import type { ReadonlyVec } from "@thi.ng/vectors";
import type { Vocab } from "./api.js";
/**
 * Jaccard distance metric wrapper for {@link kmeansDense}
 */
export declare const JACCARD_DIST_DENSE: Untransformed<ReadonlyVec>;
/**
 * k-means clustering for dense multi-hot vectors. Uses thi.ng/k-means for
 * actual clustering and squared L2 as default distance metric.
 * Default max. iterations = 100.
 *
 * @remarks
 * Use {@link JACCARD_DIST_DENSE} for alternative distance metric.
 *
 * @param k
 * @param docs
 * @param opts
 */
export declare const kmeansDense: (k: number, docs: ReadonlyVec[], opts?: Partial<KMeansOpts>) => {
    docs: ReadonlyVec[];
    id: number;
    centroid: ReadonlyVec;
    items: number[];
}[];
/**
 * k-means clustering for sparse multi-hot vectors. First converts vectors into
 * dense versions (using {@link toDense}), then calls {@link kmeansDense} to
 * perform the clustering.
 *
 * @remarks
 * Since sparse vector sizes vary, the number of dimensions used (aka the
 * vocabulary size) MUST be given via `opts`.
 *
 * @param k
 * @param docs
 * @param opts
 */
export declare const kmeansSparse: (k: number, docs: ReadonlyVec[], opts: Partial<KMeansOpts> & {
    dim: number;
}) => {
    docs: ReadonlyVec[];
    id: number;
    centroid: ReadonlyVec;
    items: number[];
}[];
export declare function clusterBounds(docs: ReadonlyVec[]): {
    centroid: ReadonlyVec;
    radius: number;
};
export declare function clusterBounds(docs: ReadonlyVec[], ids: number[]): {
    centroid: ReadonlyVec;
    radius: number;
};
/**
 * Takes a vocab and array of docs encoded as dense multi-hot vectors. Computes
 * centroid of given docs and then calls {@link centralTermsVec} to return the
 * `k`-most central terms (or less if there're insufficient non-zero vector
 * components).
 *
 * @example
 * ```ts tangle:../export/central-terms.ts
 * import { centralTerms, encodeAllDense } from "@thi.ng/text-analysis";
 *
 * const inputs = [
 *   ["a", "b", "c"],
 *   ["a", "b", "d", "e"],
 *   ["b", "f", "g"],
 *   ["a", "b", "c", "f"],
 *   ["a", "g", "h"]
 * ];
 *
 * // create vocab & encode documents into multi-hot vectors
 * const { vocab, docs } = encodeAllDense(inputs);
 *
 * // extract top-4 common terms
 * console.log(centralTerms(vocab, 4, docs));
 * // [ "b", "a", "g", "f" ]
 * ```
 *
 * @param vocab
 * @param k
 * @param docs
 */
export declare const centralTerms: (vocab: Vocab, k: number, docs: ReadonlyVec[]) => string[];
/**
 * Takes a vocab and dense vector representing a point in the n-dimensional
 * space of the given vocab. Returns an array of terms corresponding to the `k`
 * largest non-zero components of the vector (or less if there're insufficient
 * non-zero vector components).
 *
 * @remarks
 * Also see {@link centralTerms} (incl. code example).
 *
 * @param vocab
 * @param k
 * @param centroid
 */
export declare const centralTermsVec: (vocab: Vocab, k: number, centroid: ReadonlyVec) => string[];
export declare const knearest: (query: ReadonlyVec, k: number, r?: number, dist?: Untransformed<ReadonlyVec>, sorted?: boolean) => import("@thi.ng/distance").KNearest<ReadonlyVec, unknown>;
export declare const knearestDocs: (query: ReadonlyVec, k: number, docs: ReadonlyVec[], r?: number, dist?: Untransformed<ReadonlyVec>, sorted?: boolean) => [ReadonlyVec, number][];
//# sourceMappingURL=cluster.d.ts.map