faster, feature rich dijkstra for real world applications
npm install geojson-dijkstraA GeoJSON-first implementation of Dijkstra and A for NodeJS.
This repo is heavily indebted to the great ngraph.path library, which instructed a good portion of the data model for geojson-dijkstra. Extra thanks to @mourner for creating a blazing fast priority queue.
If you are looking for a more flexible, general-purpose pathfinding library, you should consider using NGraph instead.
``
npm install geojson-dijkstra --save
`
const fs = require('fs');
const { Graph, CoordinateLookup, buildGeoJsonPath, buildEdgeIdList } = require('geojson-dijkstra');
const geojson = JSON.parse(fs.readFileSync('./networks/test.geojson'));
const graph = new Graph(geojson);
// create a coordinate lookup to be able to input arbitrary coordinate pairs
// and return the nearest coordinates in the network
const lookup = new CoordinateLookup(graph);
const coords1 = lookup.getClosestNetworkPt(-101.359, 43.341);
const coords2 = lookup.getClosestNetworkPt(-91.669, 40.195);
// create a finder, in which you may specify your A* heuristic (optional)
// and add extra attributes to the result object
const finder = graph.createFinder({ parseOutputFns: [buildGeoJsonPath, buildEdgeIdList] });
// the result will contain a total_cost attribute,
// as well as additional attributes you specified when creating a finder
const result = finder.findPath(coords1, coords2);
`
Each geojson feature's properties object must contain an _id attribute (as a number) and a _cost attribute (as a non-zero number).
Additionally, the following properties can be used to customize:
* _direction: a value of "f" indicates that the linestring is valid in the forward direction only. default is valid in both directions._forward_cost
* : (number, overrides _cost) cost in the forward direction_backward_cost
* : (number, overrides _cost) cost in the backward direction
`
const graph = new Graph(geojson, options_object);
Create new new graph. options_object is an object with the following (optional) property:
mutate_inputs: (default false)
Allows the mutation of the source dataset. Dramatically speeds up loading large datasets. Do not use this options if your in-memory geojson will be used in other places in your application.
Graph Methods
`
const finder = graph.createFinder({options_object});
Creates a finder object with one property; the findPath function.
The options_object for graph.createFinder includes the following optional properties:
- heuristic: This activates A* mode, and will dramatically speed up routing. The function will be supplied with (fromCoords, toCoords), (the current and end node in a graph) in which both of parameters are of type [lng, lat]. The heuristic function will return an estimated _cost of the travel between the two nodes.
Without a heuristic function your path will be routed by a standard implementation of Dijkstra algorithm.
Defining a heuristic function that has not been properly considered can result in suboptimal paths, so beware!
The trick to creating a good heuristic function is to try to guess the _cost between the current node and the end node without overestimating.
- outputFns: These function determine the answers you will receive from your pathfinding.
You can provide a single function by itself, an array of functions, or nothing at all.
By default, running findPath will return a response with only a total_cost property:
`
{
total_cost: 123
}
To provide additional outputs, you can add (or create your own) outputFunctions functions which can parse Dijkstra internals into usable outputs.
Two built-in output functions are:
- buildGeoJsonPath
Will append { path: (geojson) } to the response object, where path is a GeoJSON linestring of all edges and properties along the shortest path.
- buildEdgeIdList
Will append { edge_list: [array, of, ids] } to the response object, where edge_list is an ordered array of edge-ids [1023, 1024, 1025], corresponding to the _id property in your input GeoJSON file.
Finder Methods
`
const path = finder.findPath(startCoordinates, endCoordinates);
Runs Dijkstra's algorithm from the startCoordinates to the endCoordinates. Coordinates must be in the form of [lng, lat], where lng and lat are both numbers (not strings!).
These coordinates must exactly correspond to network nodes in your graph (the start or end points of actual linestrings in your geoJSON). Because this can be inconvenient, the library provides a CoordinateLookup service which will take an input coordinate, and provide the closest network node.
CoordinateLookup
A fast geographically indexed coordinate lookup service leveraging geokdbush.
`
const lookup = new CoordinateLookup(graph);
const longitude = -120.868893;
const latitude = 39.500155;
const startOfPath = lookup.getClosestNetworkPt(longitude, latitude);
Provide a longitude and latitude coordinate, and receive an array: [lng, lat]` corresponding to the nearest node in your network.
Suffice to say; if this is not fast enough, you'll need to seek a solution implemented in a compiled language.
I built this library mainly as a data-structure base to build a Contraction Hierarchy.
Pre-processing a graph using a Contraction Hiererarchy can result in pathfinding performance that far exceeds A*.
Contraction Hierarchy project (in progress)