This example demonstrates accelerated two-dimensional filtering enabled by d3-quadtree. A quadtree recursively subdivides square cells into four equal-sized subcells. Each leaf node of the quadtree contains a single point. If a given quadtree cell does not intersect the brush extent, then none of the points contained in that subtree can be selected, and thus do not need to be scanned. Above, orange indicates points that are scanned but not selected. Without a quadtree, all points would need to be scanned!
Last active
April 14, 2023 11:18
-
-
Save mbostock/4343214 to your computer and use it in GitHub Desktop.
Quadtree
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| <!DOCTYPE html> | |
| <style> | |
| .point { | |
| fill: #000; | |
| fill-opacity: 0.4; | |
| } | |
| .point--scanned { | |
| fill: orange; | |
| fill-opacity: 1; | |
| stroke: orange; | |
| stroke-width: 3px; | |
| } | |
| .point--selected { | |
| fill: red; | |
| fill-opacity: 1; | |
| stroke: red; | |
| stroke-width: 5px; | |
| } | |
| .node { | |
| fill: none; | |
| stroke: #ccc; | |
| shape-rendering: crispEdges; | |
| } | |
| </style> | |
| <svg width="960" height="500"></svg> | |
| <script src="https://d3js.org/d3.v4.min.js"></script> | |
| <script> | |
| var svg = d3.select("svg"), | |
| width = +svg.attr("width"), | |
| height = +svg.attr("height"), | |
| selected; | |
| var random = Math.random, | |
| data = d3.range(2500).map(function() { return [random() * width, random() * height]; }); | |
| var quadtree = d3.quadtree() | |
| .extent([[-1, -1], [width + 1, height + 1]]) | |
| .addAll(data); | |
| var brush = d3.brush() | |
| .on("brush", brushed); | |
| svg.selectAll(".node") | |
| .data(nodes(quadtree)) | |
| .enter().append("rect") | |
| .attr("class", "node") | |
| .attr("x", function(d) { return d.x0; }) | |
| .attr("y", function(d) { return d.y0; }) | |
| .attr("width", function(d) { return d.y1 - d.y0; }) | |
| .attr("height", function(d) { return d.x1 - d.x0; }); | |
| var point = svg.selectAll(".point") | |
| .data(data) | |
| .enter().append("circle") | |
| .attr("class", "point") | |
| .attr("cx", function(d) { return d[0]; }) | |
| .attr("cy", function(d) { return d[1]; }) | |
| .attr("r", 2); | |
| svg.append("g") | |
| .attr("class", "brush") | |
| .call(brush) | |
| .call(brush.move, [[100, 100], [200, 200]]); | |
| function brushed() { | |
| var extent = d3.event.selection; | |
| point.each(function(d) { d.scanned = d.selected = false; }); | |
| search(quadtree, extent[0][0], extent[0][1], extent[1][0], extent[1][1]); | |
| point.classed("point--scanned", function(d) { return d.scanned; }); | |
| point.classed("point--selected", function(d) { return d.selected; }); | |
| } | |
| // Find the nodes within the specified rectangle. | |
| function search(quadtree, x0, y0, x3, y3) { | |
| quadtree.visit(function(node, x1, y1, x2, y2) { | |
| if (!node.length) { | |
| do { | |
| var d = node.data; | |
| d.scanned = true; | |
| d.selected = (d[0] >= x0) && (d[0] < x3) && (d[1] >= y0) && (d[1] < y3); | |
| } while (node = node.next); | |
| } | |
| return x1 >= x3 || y1 >= y3 || x2 < x0 || y2 < y0; | |
| }); | |
| } | |
| // Collapse the quadtree into an array of rectangles. | |
| function nodes(quadtree) { | |
| var nodes = []; | |
| quadtree.visit(function(node, x0, y0, x1, y1) { | |
| node.x0 = x0, node.y0 = y0; | |
| node.x1 = x1, node.y1 = y1; | |
| nodes.push(node); | |
| }); | |
| return nodes; | |
| } | |
| </script> |
Interesting revision a few days ago updating the quadtree accessor function + changing the data from maps to arrays - is that more efficient? Another potential optimization is not instantiating a few thousand svg elements outside of the view area. Just sayin'. :-)
I forked to create a little demo of finding nearest neighbor of a new point in the quadtree. See http://bl.ocks.org/patricksurry/6478178.
It's much more efficient with a data-dependent ordering of child traversal.
Also couldn't see a way of getting a node's extent (or depth) in my custom visit() routine - am I missing something?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Found initialization typo bug at line 43 - almost half the nodes are out of the view area. Fix available here: https://gist.github.com/4409108