https://github.com/rajpurkar/tsneez
Fast javascript implementation of T-SNE with tree-based acceleration
https://github.com/rajpurkar/tsneez
fast javascript t-sne
Last synced: 6 months ago
JSON representation
Fast javascript implementation of T-SNE with tree-based acceleration
- Host: GitHub
- URL: https://github.com/rajpurkar/tsneez
- Owner: rajpurkar
- License: other
- Created: 2016-11-13T05:16:48.000Z (almost 9 years ago)
- Default Branch: master
- Last Pushed: 2019-01-19T07:24:32.000Z (almost 7 years ago)
- Last Synced: 2025-03-27T21:38:55.714Z (7 months ago)
- Topics: fast, javascript, t-sne
- Language: JavaScript
- Homepage: https://rajpurkar.github.io/tsneez
- Size: 28.5 MB
- Stars: 15
- Watchers: 3
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# tsneez
## High Dimensional Visualization Simplified
### Example 1: Data passed as high dimensional pairwise dissimilar vectors
Import tsneez.js into your document: and then is some of the usage code.
```javascript
// Hyper parameters
var opt = {}
opt.theta = 0.5 // theta
opt.perplexity = 10 // perplexity
var GRADIENT_STEPS = 5000
var tsneez = new tsneez.TSNEEZ(opt) // create a tsneez instance
// Initialize data. We have four high dimensional pairwise dissimilar vectors
var vects = [
[0.418, 0.24968, -0.41242, 0.1217, 0.34527],
[0.013441, 0.23682, -0.16899, 0.40951, 0.63812],
[0.15164, 0.30177, -0.16763, 0.17684, 0.31719],
[0.70853, 0.57088, -0.4716, 0.18048, 0.54449],
]
tsneez.initData(vecs)
// A function that applies tsneez algorithm to reduce high dimenionsal vectors
function dimensionReduce(vecs) {
tsneez.initData(vecs)
for(var k = 0; k < GRADIENT_STEPS ; k++) {
tsneez.step() // gradient update
console.log(`Step : ${k}`)
}
var Y = tsneez.Y
return Y // Y is an array of 2-dimensional vectors that you can plot.
}
// Execute dimensionality reduction
dimensionReduce(vecs)
```
# Example 2: Data passed to tsneez as a set of high-dimensional Glove vectors
Append the following code within the scripts ()tags of your document.
```javascript
// Fetch data from a json file of Glove vectors and execute dimenisonal reductionality
var shortgloveFile = '/tsneez/data/shortglove.json' // A json file of Glove vectors.
document.addEventListener('DOMContentLoaded', function() {
fetch(shortgloveFile)
.then(data => data.json())
.then((gloves) => {
var vecs = gloves["vecs"]
dimensionReduce(vecs)
})
})
```
# Example 3: tsneez visualization
(a) Create a HTML document to view the visualization.
```javascript
tsneez Visualization
.container {
max-width: 800px;
margin: 0 auto;
padding: 10px;
background-color:whitesmoke;
}
.viewport {
width: 800px;
height: 600px;
}
.viewport svg {
width: 800px;
height: 600px;
display: block;
}
.viewport svg text {
pointer-events: none;
font-family: 'Lato', Arial;
font-weight: 500;
font-size: 1em;
fill: #444;
}
```
(b) Create an example3.js javascript file in the same directory as that of the HTML document and append the following code.
```javascript
// Hyper parameters
var opt = {}
opt.theta = 0.9 // theta
opt.perplexity = 20 // perplexity
function dimensionReduce(vecs) {
var NUMBER_OF_EMBEDDINGS = 1000
tsneez.initData(vecs.vecs)
visEmbedding(vecs, NUMBER_OF_EMBEDDINGS)
}
// Fetch data from a json file.
var shortgloveFile = '/tsneez/data/shortglove.json'
document.addEventListener('DOMContentLoaded', function() {
fetch(shortgloveFile)
.then(data => data.json())
.then((vecs) => {
dimensionReduce(vecs)
})
})
// define and create a tsneez instance
var stepnum = 0
var stepEmbedding, getEmbedding
var tsneez = new tsneez.TSNEEZ(opt) // create a tsneez instance
initData = function (vecs) { tsneez.initData(vecs) }
stepEmbedding = function () { stepnum++; return tsneez.step() }
getEmbedding = function () { return tsneez.Y }
var visEmbedding = function(vecs, number_of_embeddings) {
function convertToD3Format(v) {
let converted = Object.assign({}, v)
converted.words = v.words.map(function (word) {
return { str: String(word), init: true,}
})
converted = {
words: converted.words.slice(0,number_of_embeddings),
vecs: converted.vecs.slice(0,number_of_embeddings),
}
return converted
}
function animate() {
stepEmbedding()
updateEmbedding(converted.words)
window.requestAnimationFrame(animate)
}
let converted = convertToD3Format(vecs)
drawEmbedding(converted)
window.requestAnimationFrame(animate)
}
// Set up visualization
var svg
var zoomListener = d3.behavior.zoom()
.scaleExtent([0.0005, 10])
.center([0, 0])
.on('zoom', zoomHandler)
var tx = 0
var ty = 0
var ss = 1
function zoomHandler () {
tx = d3.event.translate[0]
ty = d3.event.translate[1]
ss = d3.event.scale
}
// (re-)Draw the visualization
function draw (data) {
var g = svg.selectAll('.b')
.data(data.words)//, function (d) { return d.str })
.enter().append('g')
.attr('class', 'u')
g.append('rect')
.attr('width', function (d) { return (d.str.length * 8) + 10 })
.attr('height', 20)
.attr('rx', 5)
.attr('ry', 5)
.style('fill', function (d) {
return randomColor({luminosity: 'light', seed: d.str})
})
g.append('text')
.attr('text-anchor', 'middle')
.attr('x', function (d) { return (d.str.length * 4) + 5 })
.attr('y', 10)
.attr('alignment-baseline', 'central')
.attr('fill', '#333')
.text(function (d) { return d.str })
}
// Resize the viewport in response to window resize
function resize () {
var width = $('.viewport').width()
var height = 600
svg.attr('width', width).attr('height', height)
}
// Draw initial embedding
function drawEmbedding (data) {
var div = d3.select('.viewport')
svg = div.append('svg') // svg is global
draw(data)
zoomListener(svg)
d3.select(window).on('resize', resize)
resize()
}
// Update d3 embedding on a step
function updateEmbedding (words) {
var Y = getEmbedding()
var s = svg.selectAll('.u')
.data(words)
.attr('transform', function (d, i) {
if (!d.rotate) {
d.rotate = (Math.random() - 0.5) * 10
} else {
d.rotate = d.rotate
}
return 'translate(' +
((Y.get(i, 0) * 200 * ss + tx) + 400) + ',' +
((Y.get(i, 1) * 200 * ss + ty) + 400) + ')' +
'rotate(' + d.rotate + ')'
})
}
```