StackGenVis: Alignment of Data, Algorithms, and Models for Stacking Ensemble Learning Using Performance Metrics
https://doi.org/10.1109/TVCG.2020.3030352
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
357 lines
12 KiB
357 lines
12 KiB
|
|
import KDBush from 'kdbush';
|
|
|
|
const defaultOptions = {
|
|
minZoom: 0, // min zoom to generate clusters on
|
|
maxZoom: 16, // max zoom level to cluster the points on
|
|
radius: 40, // cluster radius in pixels
|
|
extent: 512, // tile extent (radius is calculated relative to it)
|
|
nodeSize: 64, // size of the KD-tree leaf node, affects performance
|
|
log: false, // whether to log timing info
|
|
|
|
// a reduce function for calculating custom cluster properties
|
|
reduce: null, // (accumulated, props) => { accumulated.sum += props.sum; }
|
|
|
|
// properties to use for individual points when running the reducer
|
|
map: props => props // props => ({sum: props.my_value})
|
|
};
|
|
|
|
export default class Supercluster {
|
|
constructor(options) {
|
|
this.options = extend(Object.create(defaultOptions), options);
|
|
this.trees = new Array(this.options.maxZoom + 1);
|
|
}
|
|
|
|
load(points) {
|
|
const {log, minZoom, maxZoom, nodeSize} = this.options;
|
|
|
|
if (log) console.time('total time');
|
|
|
|
const timerId = `prepare ${ points.length } points`;
|
|
if (log) console.time(timerId);
|
|
|
|
this.points = points;
|
|
|
|
// generate a cluster object for each point and index input points into a KD-tree
|
|
let clusters = [];
|
|
for (let i = 0; i < points.length; i++) {
|
|
if (!points[i].geometry) continue;
|
|
clusters.push(createPointCluster(points[i], i));
|
|
}
|
|
this.trees[maxZoom + 1] = new KDBush(clusters, getX, getY, nodeSize, Float32Array);
|
|
|
|
if (log) console.timeEnd(timerId);
|
|
|
|
// cluster points on max zoom, then cluster the results on previous zoom, etc.;
|
|
// results in a cluster hierarchy across zoom levels
|
|
for (let z = maxZoom; z >= minZoom; z--) {
|
|
const now = +Date.now();
|
|
|
|
// create a new set of clusters for the zoom and index them with a KD-tree
|
|
clusters = this._cluster(clusters, z);
|
|
this.trees[z] = new KDBush(clusters, getX, getY, nodeSize, Float32Array);
|
|
|
|
if (log) console.log('z%d: %d clusters in %dms', z, clusters.length, +Date.now() - now);
|
|
}
|
|
|
|
if (log) console.timeEnd('total time');
|
|
|
|
return this;
|
|
}
|
|
|
|
getClusters(bbox, zoom) {
|
|
let minLng = ((bbox[0] + 180) % 360 + 360) % 360 - 180;
|
|
const minLat = Math.max(-90, Math.min(90, bbox[1]));
|
|
let maxLng = bbox[2] === 180 ? 180 : ((bbox[2] + 180) % 360 + 360) % 360 - 180;
|
|
const maxLat = Math.max(-90, Math.min(90, bbox[3]));
|
|
|
|
if (bbox[2] - bbox[0] >= 360) {
|
|
minLng = -180;
|
|
maxLng = 180;
|
|
} else if (minLng > maxLng) {
|
|
const easternHem = this.getClusters([minLng, minLat, 180, maxLat], zoom);
|
|
const westernHem = this.getClusters([-180, minLat, maxLng, maxLat], zoom);
|
|
return easternHem.concat(westernHem);
|
|
}
|
|
|
|
const tree = this.trees[this._limitZoom(zoom)];
|
|
const ids = tree.range(lngX(minLng), latY(maxLat), lngX(maxLng), latY(minLat));
|
|
const clusters = [];
|
|
for (const id of ids) {
|
|
const c = tree.points[id];
|
|
clusters.push(c.numPoints ? getClusterJSON(c) : this.points[c.index]);
|
|
}
|
|
return clusters;
|
|
}
|
|
|
|
getChildren(clusterId) {
|
|
const originId = clusterId >> 5;
|
|
const originZoom = clusterId % 32;
|
|
const errorMsg = 'No cluster with the specified id.';
|
|
|
|
const index = this.trees[originZoom];
|
|
if (!index) throw new Error(errorMsg);
|
|
|
|
const origin = index.points[originId];
|
|
if (!origin) throw new Error(errorMsg);
|
|
|
|
const r = this.options.radius / (this.options.extent * Math.pow(2, originZoom - 1));
|
|
const ids = index.within(origin.x, origin.y, r);
|
|
const children = [];
|
|
for (const id of ids) {
|
|
const c = index.points[id];
|
|
if (c.parentId === clusterId) {
|
|
children.push(c.numPoints ? getClusterJSON(c) : this.points[c.index]);
|
|
}
|
|
}
|
|
|
|
if (children.length === 0) throw new Error(errorMsg);
|
|
|
|
return children;
|
|
}
|
|
|
|
getLeaves(clusterId, limit, offset) {
|
|
limit = limit || 10;
|
|
offset = offset || 0;
|
|
|
|
const leaves = [];
|
|
this._appendLeaves(leaves, clusterId, limit, offset, 0);
|
|
|
|
return leaves;
|
|
}
|
|
|
|
getTile(z, x, y) {
|
|
const tree = this.trees[this._limitZoom(z)];
|
|
const z2 = Math.pow(2, z);
|
|
const {extent, radius} = this.options;
|
|
const p = radius / extent;
|
|
const top = (y - p) / z2;
|
|
const bottom = (y + 1 + p) / z2;
|
|
|
|
const tile = {
|
|
features: []
|
|
};
|
|
|
|
this._addTileFeatures(
|
|
tree.range((x - p) / z2, top, (x + 1 + p) / z2, bottom),
|
|
tree.points, x, y, z2, tile);
|
|
|
|
if (x === 0) {
|
|
this._addTileFeatures(
|
|
tree.range(1 - p / z2, top, 1, bottom),
|
|
tree.points, z2, y, z2, tile);
|
|
}
|
|
if (x === z2 - 1) {
|
|
this._addTileFeatures(
|
|
tree.range(0, top, p / z2, bottom),
|
|
tree.points, -1, y, z2, tile);
|
|
}
|
|
|
|
return tile.features.length ? tile : null;
|
|
}
|
|
|
|
getClusterExpansionZoom(clusterId) {
|
|
let clusterZoom = (clusterId % 32) - 1;
|
|
while (clusterZoom <= this.options.maxZoom) {
|
|
const children = this.getChildren(clusterId);
|
|
clusterZoom++;
|
|
if (children.length !== 1) break;
|
|
clusterId = children[0].properties.cluster_id;
|
|
}
|
|
return clusterZoom;
|
|
}
|
|
|
|
_appendLeaves(result, clusterId, limit, offset, skipped) {
|
|
const children = this.getChildren(clusterId);
|
|
|
|
for (const child of children) {
|
|
const props = child.properties;
|
|
|
|
if (props && props.cluster) {
|
|
if (skipped + props.point_count <= offset) {
|
|
// skip the whole cluster
|
|
skipped += props.point_count;
|
|
} else {
|
|
// enter the cluster
|
|
skipped = this._appendLeaves(result, props.cluster_id, limit, offset, skipped);
|
|
// exit the cluster
|
|
}
|
|
} else if (skipped < offset) {
|
|
// skip a single point
|
|
skipped++;
|
|
} else {
|
|
// add a single point
|
|
result.push(child);
|
|
}
|
|
if (result.length === limit) break;
|
|
}
|
|
|
|
return skipped;
|
|
}
|
|
|
|
_addTileFeatures(ids, points, x, y, z2, tile) {
|
|
for (const i of ids) {
|
|
const c = points[i];
|
|
const f = {
|
|
type: 1,
|
|
geometry: [[
|
|
Math.round(this.options.extent * (c.x * z2 - x)),
|
|
Math.round(this.options.extent * (c.y * z2 - y))
|
|
]],
|
|
tags: c.numPoints ? getClusterProperties(c) : this.points[c.index].properties
|
|
};
|
|
const id = c.numPoints ? c.id : this.points[c.index].id;
|
|
if (id !== undefined) {
|
|
f.id = id;
|
|
}
|
|
tile.features.push(f);
|
|
}
|
|
}
|
|
|
|
_limitZoom(z) {
|
|
return Math.max(this.options.minZoom, Math.min(z, this.options.maxZoom + 1));
|
|
}
|
|
|
|
_cluster(points, zoom) {
|
|
const clusters = [];
|
|
const {radius, extent, reduce} = this.options;
|
|
const r = radius / (extent * Math.pow(2, zoom));
|
|
|
|
// loop through each point
|
|
for (let i = 0; i < points.length; i++) {
|
|
const p = points[i];
|
|
// if we've already visited the point at this zoom level, skip it
|
|
if (p.zoom <= zoom) continue;
|
|
p.zoom = zoom;
|
|
|
|
// find all nearby points
|
|
const tree = this.trees[zoom + 1];
|
|
const neighborIds = tree.within(p.x, p.y, r);
|
|
|
|
let numPoints = p.numPoints || 1;
|
|
let wx = p.x * numPoints;
|
|
let wy = p.y * numPoints;
|
|
|
|
let clusterProperties = reduce && numPoints > 1 ? this._map(p, true) : null;
|
|
|
|
// encode both zoom and point index on which the cluster originated
|
|
const id = (i << 5) + (zoom + 1);
|
|
|
|
for (const neighborId of neighborIds) {
|
|
const b = tree.points[neighborId];
|
|
// filter out neighbors that are already processed
|
|
if (b.zoom <= zoom) continue;
|
|
b.zoom = zoom; // save the zoom (so it doesn't get processed twice)
|
|
|
|
const numPoints2 = b.numPoints || 1;
|
|
wx += b.x * numPoints2; // accumulate coordinates for calculating weighted center
|
|
wy += b.y * numPoints2;
|
|
|
|
numPoints += numPoints2;
|
|
b.parentId = id;
|
|
|
|
if (reduce) {
|
|
if (!clusterProperties) clusterProperties = this._map(p, true);
|
|
reduce(clusterProperties, this._map(b));
|
|
}
|
|
}
|
|
|
|
if (numPoints === 1) {
|
|
clusters.push(p);
|
|
} else {
|
|
p.parentId = id;
|
|
clusters.push(createCluster(wx / numPoints, wy / numPoints, id, numPoints, clusterProperties));
|
|
}
|
|
}
|
|
|
|
return clusters;
|
|
}
|
|
|
|
_map(point, clone) {
|
|
if (point.numPoints) {
|
|
return clone ? extend({}, point.properties) : point.properties;
|
|
}
|
|
const original = this.points[point.index].properties;
|
|
const result = this.options.map(original);
|
|
return clone && result === original ? extend({}, result) : result;
|
|
}
|
|
}
|
|
|
|
function createCluster(x, y, id, numPoints, properties) {
|
|
return {
|
|
x, // weighted cluster center
|
|
y,
|
|
zoom: Infinity, // the last zoom the cluster was processed at
|
|
id, // encodes index of the first child of the cluster and its zoom level
|
|
parentId: -1, // parent cluster id
|
|
numPoints,
|
|
properties
|
|
};
|
|
}
|
|
|
|
function createPointCluster(p, id) {
|
|
const [x, y] = p.geometry.coordinates;
|
|
return {
|
|
x: lngX(x), // projected point coordinates
|
|
y: latY(y),
|
|
zoom: Infinity, // the last zoom the point was processed at
|
|
index: id, // index of the source feature in the original input array,
|
|
parentId: -1 // parent cluster id
|
|
};
|
|
}
|
|
|
|
function getClusterJSON(cluster) {
|
|
return {
|
|
type: 'Feature',
|
|
id: cluster.id,
|
|
properties: getClusterProperties(cluster),
|
|
geometry: {
|
|
type: 'Point',
|
|
coordinates: [xLng(cluster.x), yLat(cluster.y)]
|
|
}
|
|
};
|
|
}
|
|
|
|
function getClusterProperties(cluster) {
|
|
const count = cluster.numPoints;
|
|
const abbrev =
|
|
count >= 10000 ? `${Math.round(count / 1000) }k` :
|
|
count >= 1000 ? `${Math.round(count / 100) / 10 }k` : count;
|
|
return extend(extend({}, cluster.properties), {
|
|
cluster: true,
|
|
cluster_id: cluster.id,
|
|
point_count: count,
|
|
point_count_abbreviated: abbrev
|
|
});
|
|
}
|
|
|
|
// longitude/latitude to spherical mercator in [0..1] range
|
|
function lngX(lng) {
|
|
return lng / 360 + 0.5;
|
|
}
|
|
function latY(lat) {
|
|
const sin = Math.sin(lat * Math.PI / 180);
|
|
const y = (0.5 - 0.25 * Math.log((1 + sin) / (1 - sin)) / Math.PI);
|
|
return y < 0 ? 0 : y > 1 ? 1 : y;
|
|
}
|
|
|
|
// spherical mercator to longitude/latitude
|
|
function xLng(x) {
|
|
return (x - 0.5) * 360;
|
|
}
|
|
function yLat(y) {
|
|
const y2 = (180 - y * 360) * Math.PI / 180;
|
|
return 360 * Math.atan(Math.exp(y2)) / Math.PI - 90;
|
|
}
|
|
|
|
function extend(dest, src) {
|
|
for (const id in src) dest[id] = src[id];
|
|
return dest;
|
|
}
|
|
|
|
function getX(p) {
|
|
return p.x;
|
|
}
|
|
function getY(p) {
|
|
return p.y;
|
|
}
|
|
|