Skip to content

Instantly share code, notes, and snippets.

@hardbyte

hardbyte/README.md

Last active April 22, 2018 13:24
Show Gist options
  • Save hardbyte/40cd6622cffbe98055d3 to your computer and use it in GitHub Desktop.
Save hardbyte/40cd6622cffbe98055d3 to your computer and use it in GitHub Desktop.
Download ZIP
PCA with D3js
Raw
index.html
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<!-- Latest compiled and minified CSS -->
<link rel="stylesheet" href="//maxcdn.bootstrapcdn.com/bootstrap/3.2.0/css/bootstrap.min.css">
<!-- Optional theme -->
<link rel="stylesheet" href="//maxcdn.bootstrapcdn.com/bootstrap/3.2.0/css/bootstrap-theme.min.css">
</head>
<body>
<div class="container-fluid">
<div class="row">
<div class="col-sm-4">
<h3>Covariance Matrix</h3>
<pre id="covar"></pre>
<h3>Eigenvectors</h3>
<pre id="eigvects"></pre>
<h3>Data Generation</h3>
<form role="form" class="form">
<div class="row">
<div class="col-md-6">
<h5>A</h5>
<div class="form-group">
<label for="amean">mean: </label>
<input class="form-control" type="number" min="0.0" max="1.0" step="0.01" width="4" id="amean" value="0.6"/>
</div>
<div class="form-group">
<label for="astd">std: </label>
<input class="form-control" type="number" min="0.0" max="1.0" step="0.01" width="4" id="astd" value="0.05"/>
</div>
<hr>
<div class="form-group">
<label for="ax">ax: </label>
<input class="form-control" type="number" min="0.0" max="1.0" step="0.01" width="4" id="ax" value="0.65"/>
</div>
<div class="form-group">
<label for="ax">ay: </label>
<input class="form-control" type="number" min="0.0" max="1.0" step="0.01" width="4" id="ay" value="0.20"/>
</div>
</div>
<div class="col-md-6">
<h5>B</h5>
<div class="form-group">
<label for="bmean">mean: </label>
<input class="form-control" type="number" min="0.0" max="1.0" step="0.01" width="4" id="bmean" value="0.2"/>
</div>
<div class="form-group">
<label for="bstd">std: </label>
<input class="form-control" type="number" min="0.0" max="1.0" step="0.01" width="4" id="bstd" value="0.08"/>
</div>
<hr>
<div class="form-group">
<label for="ax">bx: </label>
<input class="form-control" type="number" min="0.0" max="1.0" step="0.01" width="4" id="bx" value="0.35"/>
</div>
<div class="form-group">
<label for="ax">by: </label>
<input class="form-control" type="number" min="0.0" max="1.0" step="0.01" width="4" id="by" value="0.8"/>
</div>
</div>
</div>
</form>
<pre>
x = ax * a + by * b
y = ay * a + by * b</pre>
<button onclick="createData()">Generate new Data</button>
</div>
<div class="col-sm-8">
<div id="vis" class="row"></div>
<form class="">
<input type="radio" name="ax" id="ax-both"/>
<label for="ax-both">Both eigenvectors</label>
<input type="radio" name="ax" id="ax-orig"/>
<label for="ax-orig">Original vectors</label>
<br/>
<input type="radio" name="ax" id="ax-first"/>
<label for="ax-first">First eigenvectors</label>
<input type="radio" name="ax" id="ax-second"/>
<label for="ax-second">Second eigenvectors</label>
</form>
</div>
</div>
</div>
<!-- Latest compiled and minified Bootstap JavaScript -->
<script src="https://ajax.googleapis.com/ajax/libs/jquery/1.11.1/jquery.min.js"></script>
<script src="//maxcdn.bootstrapcdn.com/bootstrap/3.2.0/js/bootstrap.min.js"></script>
<script src="http://d3js.org/d3.v3.min.js"></script>
<script src="http://numericjs.com/lib/numeric-1.2.6.min.js"></script>
<script>
var margin = {top: 20, right: 20, bottom: 30, left: 50},
width = 550 - margin.left - margin.right,
height = 550 - margin.top - margin.bottom;
var X = d3.scale.linear()
.range([0, width]);
var Y = d3.scale.linear()
.range([height, 0]);
var svg = d3.select("#vis").append("svg")
.attr("width", width + margin.left + margin.right)
.attr("height", height + margin.top + margin.bottom)
.append("g")
.attr("transform", "translate(" + margin.left + "," + margin.top + ")");
var numSamplesPerFrame = 200,
numSamples = 0;
var xdata = [],
ydata = [];
var x_mean, y_mean;
var data, eigs;
// Event binding
d3.selectAll("input[name=ax]")
.data([[true, true], [false, false], [true, false], [false, true]])
.on("change", function(d) {
move(d[0], d[1]);
});
// Draw the normal axes
var components = svg.selectAll("line")
.data([
[[0.0, 0.5], [1, 0.5], "X"],
[[0.5, 0.0], [0.5, 1], "Y"],
], function(d, i){return d[2];});
components.exit().remove();
components.enter().append('line')
.attr('class', 'ax')
.attr('stroke', 'black')
.attr('stroke-width', '2px')
.attr('x1', function (d) { return X(d[0][0]); })
.attr('y1', function (d) { return Y(d[0][1]); })
.attr('x2', function (d) { return X(d[1][0]); })
.attr('y2', function (d) { return Y(d[1][1]); });
function rnd(mean, std){
var r = 0;
for (var i = 0; i < 10; i++) {
r += Math.random() * 2 - 1
}
return r * std + mean;
}
function covariance(x_adjust, y_adjust, n){
var total = 0;
for (var i = 0; i < x_adjust.length; i++) {
total += x_adjust[i] * y_adjust[i];
}
return total / (n-1);
}
function solve() {
var n = xdata.length;
x_mean = xdata.reduce(function (memo, num) {
return memo + num;
}, 0) / n;
y_mean = ydata.reduce(function (memo, num) {
return memo + num;
}, 0) / n;
// Subtract the mean
var x_adjust = xdata.map(function (num) {
return num - x_mean;
});
var y_adjust = ydata.map(function (num) {
return num - y_mean;
});
data = [y_adjust, x_adjust];
// Calculate the covariance
var xyc = covariance(x_adjust, y_adjust, n);
var xxc = covariance(x_adjust, x_adjust, n);
var yyc = covariance(y_adjust, y_adjust, n);
var covar = [ [xxc, xyc], [xyc, yyc] ];
d3.selectAll('#covar').html(numeric.prettyPrint(covar));
// Calculate the eigen vectors
var eig = numeric.eig(covar);
eigs = eig.E.x;
var eigelements = d3.select("#eigvects").selectAll("div").data(eigs);
eigelements.enter().append('div')
.html(function(d){return numeric.prettyPrint(d)});
}
function plotEigVects(){
// Plot eigen vectors centered at the mean
var components = svg.selectAll("line.pca")
.data([
[
[x_mean, y_mean],
[x_mean + eigs[0][1], y_mean + eigs[0][0]],
"Y"
],
[
[x_mean, y_mean],
[x_mean + eigs[1][1], y_mean + eigs[1][0]],
"X"
]
], function(d){return d[2];});
components.enter().append('line')
.attr('stroke', 'red')
.attr('stroke-width', '2px')
.attr("class", 'pca');
components
.transition().ease("linear").duration(2000)
.attr('x1', function (d) { return X(d[0][0]); })
.attr('y1', function (d) { return Y(d[0][1]); })
.attr('x2', function (d) { return X(d[1][0]); })
.attr('y2', function (d) { return Y(d[1][1]); });
}
function move(useFirst, useSecond){
// Form a feature vector
var featureVectorRow = [];
if(useFirst){
featureVectorRow.push(eigs[0]);
}
if(useSecond){
featureVectorRow.push(eigs[1]);
}
if(!useFirst && !useSecond){
// plot the original data
plotOriginal();
plotEigVects();
return;
}
var finalData = numeric.transpose(numeric.dot(featureVectorRow, data));
var n = xdata.length;
// Move the axis lines to normal positions
var components = svg.selectAll("line.pca")
.data([
[[0.5, 0.5], [1, 0.5], "X"],
[[0.5, 0.5], [0.5, 1], "Y"],
], function(d, i){return d[2];});
components.exit().remove();
components.enter().append('line');
// plot the new lower dimensional data
var circle = svg.selectAll("circle");
//.data(finalData, function(d, i){ return d;});
circle
.transition().ease("linear").duration(2000)
.attr('cx', function(d, i){
if(useFirst && useSecond) {
return X(0.5 + finalData[i][1]);
}
if(useSecond) {
return X(0.5 + finalData[i][0]);
}
if (useFirst){
return X(0.5);//X(i/n);
}
})
.attr('cy', function(d, i){
if(useFirst && useSecond) {
return Y(0.5 + finalData[i][0]);
}
if (useFirst){
return Y(0.5 + finalData[i][0]);
}
if (useSecond){
return Y(0.5);
}
});
components
.transition().ease("linear").duration(2000)
.attr('x1', function (d) { return X(d[0][0]); })
.attr('y1', function (d) { return Y(d[0][1]); })
.attr('x2', function (d) { return X(d[1][0]); })
.attr('y2', function (d) { return Y(d[1][1]); });
}
function plotOriginal(){
var xycoords = numeric.transpose([xdata, ydata]);
var circle = svg.selectAll("circle")
.data(xycoords, function(d, i) { return d; });
circle.enter().append("circle")
.attr("r", 2)
.attr("fill", 'steelblue');
circle
.transition().ease("linear").duration(2000)
.attr("cx", function(d, i) { return X(d[0]); })
.attr("cy", function(d, i){return Y(d[1]);});
circle.exit().remove();
}
function createData() {
eigs = [];
xdata = [];
ydata = [];
numSamples = 0;
var amean = parseFloat(document.getElementById('amean').value);
var astd = parseFloat(document.getElementById('astd').value);
var bmean = parseFloat(document.getElementById('bmean').value);
var bstd = parseFloat(document.getElementById('bstd').value);
var ax = parseFloat(document.getElementById('ax').value);
var ay = parseFloat(document.getElementById('ay').value);
var bx = parseFloat(document.getElementById('bx').value);
var by = parseFloat(document.getElementById('by').value);
d3.timer(function () {
for (var i = 0; i < numSamplesPerFrame; ++i) {
var a = rnd(amean, astd),
b = rnd(bmean, bstd);
var x = ax * a + bx * b,
y = ay * a + by * b;
xdata.push(x);
ydata.push(y);
}
plotOriginal();
if (++numSamples > 10) {
solve();
plotEigVects();
return true;
}
});
}
createData();
</script>
</body>
</html>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment