[!NOTE] This is one of 190 standalone projects, maintained as part of the @thi.ng/umbrella monorepo and anti-framework.
🚀 Please help me to work full-time on these projects by sponsoring me on GitHub. Thank you! ❤️
N-dimensional distance metrics & K-nearest neighborhoods for point queries.
The package provides the IDistance
interface for custom distance metric implementations & conversions from/to raw distance values. The following preset metrics are provided too:
Preset | Number | nD | 2D | 3D | Comments |
---|---|---|---|---|---|
EUCLEDIAN | ✅ | Eucledian distance | |||
EUCLEDIAN1 | ✅ | ||||
EUCLEDIAN2 | ✅ | ||||
EUCLEDIAN3 | ✅ | ||||
HAVERSINE_LATLON | ✅ | Great-circle distance for lat/lon geo locations | |||
HAVERSINE_LONLAT | ✅ | Great-circle distance for lon/lat geo locations | |||
DIST_SQ | ✅ | Squared dist (avoids Math.sqrt ) | |||
DIST_SQ1 | ✅ | ||||
DIST_SQ2 | ✅ | ||||
DIST_SQ3 | ✅ | ||||
defManhattan(n) | ✅ | Manhattan distance | |||
MANHATTAN2 | ✅ | ||||
MANHATTAN3 | ✅ |
Neighborhoods can be used to select n-D spatial items around a given target location and an optional catchment radius (infinite by default). Neighborhoods also use one of the given distance metrics and implement the widely used IDeref
interface to obtain the final query results.
Custom neighborhood selections can be defined via the INeighborhood
interface. Currently, there are two different implementations available, each providing several factory functions to instantiate and provide defaults for different dimensions. See documentation and examples below.
An INeighborhood
implementation for nearest neighbor queries around a given target location, initial query radius and IDistance
metric to determine proximity.
An INeighborhood
implementation for K-nearest neighbor queries around a given target location, initial query radius and IDistance
metric to determine proximity. The K-nearest neighbors will be accumulated via an internal heap and results can be optionally returned in order of proximity (via .deref()
or .values()
). For K=1 it will be more efficient to use Nearest
to avoid the additional overhead.
An unbounded and unsorted version of KNearest
, selecting all items around the target location and given search radius. Qualifying neighbors will be accumulated in order of processing via an internal array.
STABLE - used in production
Search or submit any issues for this package
Work is underway integrating this approach into the spatial indexing data structures provided by the @thi.ng/geom-accel package.
yarn add @thi.ng/distance
ES module import:
<script type="module" src="https://cdn.skypack.dev/@thi.ng/distance"></script>
For Node.js REPL:
const distance = await import("@thi.ng/distance");
Package sizes (brotli'd, pre-treeshake): ESM: 1.38 KB
One project in this repo's /examples directory is using this package:
Screenshot | Description | Live demo | Source |
---|---|---|---|
K-nearest neighbor search in an hash grid | Demo | Source |
import * as d from "@thi.ng/distance";
const items = { a: 5, b: 16, c: 9.5, d: 2, e: 12 };
// collect the 3 nearest numbers for target=10 and using
// infinite selection radius and squared distance metric (defaults)
const k = d.knearestN(10, 3);
// consider each item for inclusion
Object.entries(items).forEach(([id, x]) => k.consider(x, id));
// retrieve result tuples of [distance, value]
k.deref()
// [ [ 25, 'a' ], [ 4, 'e' ], [ 0.25, 'c' ] ]
// result values only
k.values()
// [ 'a', 'e', 'c' ]
// neighborhood around 10, K=3 w/ max radius 5
// also use Eucledian distance and sort results by proximity
const k2 = d.knearestN(10, 3, 5, d.EUCLEDIAN1, true);
Object.entries(items).forEach(([id, x]) => k2.consider(x, id));
k2.deref()
// [ [ 0.5, 'c' ], [ 2, 'e' ], [ 5, 'a' ] ]
If this project contributes to an academic publication, please cite it as:
@misc{thing-distance,
title = "@thi.ng/distance",
author = "Karsten Schmidt",
note = "https://thi.ng/distance",
year = 2021
}
© 2021 - 2024 Karsten Schmidt // Apache License 2.0
Generated using TypeDoc