DruidJS workerNatural breaksWord Tour: 40k words and their friendsHead/tail breaksHello, kmeans-engineHello, jsgeoda!n-dimensions binning?Ternary binningHello, Reorder.jsGeometric MedianUMAP SleepwalkDisc TransportSelf-Organizing Maps meet DelaunayHello, AutoencoderCumulative binningPlateau detectionSpherical smallest-circle problemFrida Kahlo’s paletteMean shift clustering with flatbushMean shift clusteringMean shift clustering with kd-treeManhattan Voronoi IIH3 odditiesProtein MatrixUMAP in A-FrameUMAP of temperatures in Minneapolis/Saint-PaulUMAP Simple TopologiesUMAP & a personalized distance functiont-SNE VoronoiTravelling salesperson approximation with t-SNEtsne.js & workertsne.jsHello tfjs-tsneTravelling salesperson approximation on the globe, with UMAP-jsUMAP-js, worker versionHello UMAP (via Algorithmia)MNIST & UMAP-jsHello UMAP-js
UMAP-o-Matic
Also listed in…
Algorithms
vectors = {
const vectors = data.map(d => numericfields.map(f => d[f]));
if (normalize) {
for (let i = 0; i < numericfields.length; i++) {
const mean = d3.mean(data, d => d[i]),
deviation = d3.deviation(data, d => d[i]),
scale = d3.scaleSymlog();
if (deviation > 0)
for (const d of vectors) d[i] = scale((d[i] - mean) / deviation);
}
}
return vectors.slice(0, maxpoints);
}
config = ({
nComponents: +nComponents,
minDist: +minDist,
nNeighbors: +nNeighbors,
nEpochs: nEpochs === "default" ? undefined : +nEpochs
})
color = d3
.scaleLinear()
.range(["red", "lime"])
.domain([0, vectors.length])
dynamic = true
// want static rendering and no worker? see #UMAP below
/*
* Use this block of code to develop your distance functions and test them synchronously.
* Then paste it into the ‘distances’ cell below, that is actually used in the worker
*/
D = {
const D = {
one: (a, b) => 1,
random: (a, b) => Math.random(),
euclidian: (a, b) =>
Math.sqrt(a.map((_, i) => (a[i] - b[i]) ** 2).reduce((a, b) => a + b, 0)),
cosine: (a, b) => {
const dot = a.map((_, i) => a[i] * b[i]).reduce((a, b) => a + b, 0),
norm2A = a.map(u => u ** 2).reduce((a, b) => a + b, 0),
norm2B = norm2A ? b.map(u => u ** 2).reduce((a, b) => a + b, 0) : 0,
N = norm2A * norm2B;
return !N ? 1 : Math.acos(dot / Math.sqrt(N)) / Math.PI;
}
};
return D;
}
function distances({ vectors, distanceType }) {
const D = {
one: (a, b) => 1,
random: (a, b) => Math.random(),
euclidian: (a, b) =>
Math.sqrt(a.map((_, i) => (a[i] - b[i]) ** 2).reduce((a, b) => a + b, 0)),
cosine: (a, b) => {
const dot = a.map((_, i) => a[i] * b[i]).reduce((a, b) => a + b, 0),
norm2A = a.map(u => u ** 2).reduce((a, b) => a + b, 0),
norm2B = norm2A ? b.map(u => u ** 2).reduce((a, b) => a + b, 0) : 0,
N = norm2A * norm2B;
return !N ? 1 : Math.acos(dot / Math.sqrt(N)) / Math.PI;
}
};
const distance = D[distanceType] || D["euclidian"];
const N = vectors.length,
distances = Float32Array.from({ length: N ** 2 });
for (let i = 0; i < N; i++) {
const a = vectors[i];
distances[i + N * i] = 0;
for (let j = 0; j < i; j++) {
const b = vectors[j];
distances[i + N * j] = distances[j + N * i] = distance(a, b);
}
}
return distances;
}
precomputed = Generators.observe(worker(distances, { vectors, distanceType }))
res = Generators.observe(
worker(
dynamic ? fitdynamic : fit,
{ distances: precomputed, labels, config },
`
const window = {};
importScripts("https://unpkg.com/umap-js@1.3.1/lib/umap-js.js");
const UMAP = window.UMAP;
`
)
)
function fit({ distances, labels, config }) {
const time = performance.now();
const umap = new UMAP(config);
if (labels) umap.setSupervisedProjection(labels);
const flat = typeof distances[0] === "number",
N = flat ? Math.sqrt(distances.length) : distances.length;
umap.distanceFn = flat
? function(i, j) {
return distances[i + N * j];
}
: function(i, j) {
return distances[i] ? distances[i][j] : 1000;
};
return {
result: umap.fit(Uint32Array.from({ length: N }, (_, i) => i)),
time: performance.now() - time
};
}
function* fitdynamic({ distances, labels, config }) {
const time = performance.now();
const umap = new UMAP(config);
if (labels) umap.setSupervisedProjection(labels);
yield {
result: null,
time: performance.now() - time,
epoch: "computing k-nearest neighbors"
};
const flat = typeof distances[0] === "number",
N = flat ? Math.sqrt(distances.length) : distances.length;
umap.distanceFn = flat
? function(i, j) {
return distances[i + N * j];
}
: function(i, j) {
return distances[i] ? distances[i][j] : 1000;
};
const nEpochs = umap.initializeFit(
Uint32Array.from({ length: N }, (_, i) => i)
);
var t = performance.now(),
t0 = t;
for (let i = 0; i < nEpochs; i++) {
umap.step();
t = performance.now();
if (t - t0 > 200) {
t0 = t;
yield {
result: umap.getEmbedding(),
time: performance.now() - time,
epoch: i
};
}
}
yield {
result: umap.getEmbedding(),
time: performance.now() - time,
epoch: nEpochs
};
}
// This symbol is not used, because UMAP is loaded in the worker
UMAP = "🌶"
// redefine if you want to try a non-worker version, with:
/*
res = fit({ distances: precomputed, labels, config })
UMAP = await require("https://unpkg.com/umap-js@1.3.0/lib/umap-js.js").catch(
() => window.UMAP
)
*/
d3 = require("d3@5")
mutable loaded = ""
loader = {
switch (testcsv) {
case "Iris":
return require("@observablehq/iris")
.then(l => l.map(d => ((d.color = classify(d.species)), d)))
.then(d => (mutable loaded = d));
case "Diamonds":
return require("@observablehq/diamonds")
.then(l => l.map(d => ((d.color = classify(d.cut)), d)))
.then(d => (mutable loaded = d));
default:
return (
d3
.text(
"https://raw.githubusercontent.com/codebrainz/color-names/master/output/colors.csv"
)
.then(t => d3.csvParse("code,name,color,r,g,b\n" + t))
// cheap labelling
.then(t =>
t.map(d => {
const m = d.code.match(/red|green|blue|yellow|orange|pink/);
if (m) d.base = m[0];
else d.base = "";
return d;
})
)
.then(d => (mutable loaded = d))
);
}
}
One platform to build and deploy the best data apps
Experiment and prototype by building visualizations in live JavaScript notebooks. Collaborate with your team and decide which concepts to build out.
Use Observable Framework to build data apps locally. Use data loaders to build in any language or library, including Python, SQL, and R.
Seamlessly deploy to Observable. Test before you ship, use automatic deploy-on-commit, and ensure your projects are always up-to-date.