Research Application Manager with the Alan Turing Institute / PhD from @GC_CUNY / formerly Research Software Engineer @LivingwMachines and @dhinstitutes project mgr
Public
Edited
Jul 11, 2023
sentences = testData
topics = getTopics(sentences, {
returnType: "object",
stemmer: false,
topics: 5
})
topics2 = getTopics(sentences, {
topics: 7, // generate 7 topics
minLength: 10, // only include words longer than or equal to 10 characters
stopwords: [], // skip stopwords
returnType: "object" // return an object
})
{
const vocabulary = topics2.vocab;
return html`<li>${vocabulary.join("<li>")}`;
}
topics3 = getTopics(sentences, {
topics: 7, // generate 7 topics
minLength: 10, // only include words longer than or equal to 10 characters
stopwords: [], // skip stopwords
returnType: "html" // return the **HTML table**
})
getTopics = (
sentences,
{
topics = 4,
stopwords = autoStopwords,
returnType = "object",
minLength = 2,
stemmer = autoStemmer
} = {}
) => {
var documents = new Array(),
f = {},
vocab = new Array(),
docCount = 0;
if (!Array.isArray(sentences)) sentences = sentences.split("\n");
sentences = sentences
// remove links
.map((sentence) => sentence.replace(/(?:https?|ftp):\/\/[\n\S]+/g, ""))
// remove characters
.map((sentence) => sentence.replace(/[^a-z\'A-Z ]+/g, " "))
.map((sentence) => sentence.trim())
.map((sentence) => sentence.toLowerCase())
.filter((d) => d);
sentences.forEach((sentence) => {
var words = sentence.split(/[\s,\"]+/);
if (!words) return;
var wordIndices = new Array();
words = words
.filter((word) => word)
.filter((word) => !stopwords[word])
.filter((word) => word.length >= minLength);
if (stemmer) words = words.map((word) => stemmer(word));
words.forEach((word) => {
if (f[word]) {
f[word] = f[word] + 1;
} else if (word) {
f[word] = 1;
vocab.push(word);
}
wordIndices.push(vocab.indexOf(word));
});
if (wordIndices && wordIndices.length > 0) {
documents[docCount++] = wordIndices;
}
});
var V = vocab.length,
M = documents.length,
K = topics,
alpha = 0.1, // per-document distributions over topics
beta = 0.01; // per-topic distributions over words
lda.configure(documents, V, 10000, 2000, 100, 10);
lda.gibbs(K, alpha, beta);
var theta = lda.getTheta(),
phi = lda.getPhi();
var htmlOutput = "",
data = {
topics: {},
vocab,
sentences,
documents
};
//topics
var topicText = new Array(),
topTerms = 20;
phi.forEach((p, pix) => {
topicText[pix] = "";
var tuples = new Array();
p.forEach((val, ix) => {
tuples.push("" + val.toPrecision(2) + "_" + vocab[ix]);
});
tuples.sort().reverse();
if (topTerms > vocab.length) topTerms = vocab.length;
data.topics[pix] = [];
Array(topTerms)
.fill(0)
.forEach((d, t) => {
var topicTerm = tuples[t].split("_")[1];
var prob = parseInt(tuples[t].split("_")[0] * 100);
if (prob < 0.0001) return;
data.topics[pix].push([topicTerm, prob]);
topicText[pix] += topicTerm + " ";
});
topicText[pix] = topicText[pix].trim();
});
data.topicText = topicText;
htmlOutput = '<div class="spacer"> </div>';
theta.forEach((t, ix) => {
htmlOutput += '<div class="lines">';
htmlOutput +=
'<div style="display:table-cell;width:100px;padding-right:5px">';
t.forEach((k, tix) => {
const width = parseInt("" + k * 100);
htmlOutput += `<div class="box bgcolor${tix}" style="width:${width}px" title="${topicText[tix]}"></div>`;
});
htmlOutput += `</div>${sentences[ix]}</div>`;
});
data.theta = theta;
if (returnType === "object") return data;
const style = `<style>
.box{
height: 8px;
vertical-align: middle;
margin-top: 8px;
display: table-cell;
}
.color0{ color:#FF0000;} .bgcolor0{ background-color:#FF0000;}
.color1{ color:#0000FF;} .bgcolor1{ background-color:#0000FF;}
.color2{ color:#00FF00;} .bgcolor2{ background-color:#00FF00;}
.color3{ color:#808000;} .bgcolor3{ background-color:#808000;}
.color4{ color:#00FF00;} .bgcolor4{ background-color:#00FF00;}
.color5{ color:#008000;} .bgcolor5{ background-color:#008000;}
.color6{ color:#00FFFF;} .bgcolor6{ background-color:#00FFFF;}
.color7{ color:orange;} .bgcolor7{ background-color:orange;}
.color8{ color:#800000;} .bgcolor8{ background-color:#800000;}
.color9{ color:#000080;} .bgcolor9{ background-color:#000080;}
.color10{ color:#FF00FF;} .bgcolor10{ background-color:#FF00FF;}
.color11{ color:#800080;} .bgcolor11{ background-color:#800080;}
.color12{ color:#C0C0C0;} .bgcolor12{ background-color:#C0C0C0;}
.lines {
margin: 0px;
padding: 2px;
display: table;
}
</style>
`;
return html`${style}${htmlOutput}`;
}
lda = new (function () {
var documents,
z,
nw,
nd,
nwsum,
ndsum,
thetasum,
phisum,
V,
K,
alpha,
beta,
THIN_INTERVAL = 20,
BURN_IN = 100,
ITERATIONS = 1000,
SAMPLE_LAG,
dispcol = 0,
numstats = 0;
this.configure = function (
docs,
v,
iterations,
burnIn,
thinInterval,
sampleLag
) {
this.ITERATIONS = iterations;
this.BURN_IN = burnIn;
this.THIN_INTERVAL = thinInterval;
this.SAMPLE_LAG = sampleLag;
this.documents = docs;
this.V = v;
this.dispcol = 0;
this.numstats = 0;
};
this.initialState = function (K) {
var i,
M = this.documents.length;
this.nw = make2DArray(this.V, K);
this.nd = make2DArray(M, K);
this.nwsum = makeArray(K);
this.ndsum = makeArray(M);
this.z = new Array();
for (i = 0; i < M; i++) this.z[i] = new Array();
for (var m = 0; m < M; m++) {
var N = this.documents[m].length;
this.z[m] = new Array();
for (var n = 0; n < N; n++) {
var topic = parseInt("" + Math.random() * K);
this.z[m][n] = topic;
this.nw[this.documents[m][n]][topic]++;
this.nd[m][topic]++;
this.nwsum[topic]++;
}
this.ndsum[m] = N;
}
};
this.gibbs = function (K, alpha, beta) {
var i;
this.K = K;
this.alpha = alpha;
this.beta = beta;
if (this.SAMPLE_LAG > 0) {
this.thetasum = make2DArray(this.documents.length, this.K);
this.phisum = make2DArray(this.K, this.V);
this.numstats = 0;
}
this.initialState(K);
console.log(
`Sampling ${this.ITERATIONS} iterations with burn-in of ${this.BURN_IN} (B/S=${this.THIN_INTERVAL}).`
);
for (i = 0; i < this.ITERATIONS; i++) {
for (var m = 0; m < this.z.length; m++) {
for (var n = 0; n < this.z[m].length; n++) {
var topic = this.sampleFullConditional(m, n);
this.z[m][n] = topic;
}
}
if (i < this.BURN_IN && i % this.THIN_INTERVAL == 0) {
this.dispcol++;
}
if (i > this.BURN_IN && i % this.THIN_INTERVAL == 0) {
this.dispcol++;
}
if (i > this.BURN_IN && this.SAMPLE_LAG > 0 && i % this.SAMPLE_LAG == 0) {
this.updateParams();
if (i % this.THIN_INTERVAL != 0) this.dispcol++;
}
if (this.dispcol >= 100) {
this.dispcol = 0;
}
}
};
this.sampleFullConditional = function (m, n) {
var topic = this.z[m][n];
this.nw[this.documents[m][n]][topic]--;
this.nd[m][topic]--;
this.nwsum[topic]--;
this.ndsum[m]--;
var p = makeArray(this.K);
for (var k = 0; k < this.K; k++) {
p[k] =
(((this.nw[this.documents[m][n]][k] + this.beta) /
(this.nwsum[k] + this.V * this.beta)) *
(this.nd[m][k] + this.alpha)) /
(this.ndsum[m] + this.K * this.alpha);
}
for (var k = 1; k < p.length; k++) {
p[k] += p[k - 1];
}
var u = Math.random() * p[this.K - 1];
for (topic = 0; topic < p.length; topic++) {
if (u < p[topic]) break;
}
this.nw[this.documents[m][n]][topic]++;
this.nd[m][topic]++;
this.nwsum[topic]++;
this.ndsum[m]++;
return topic;
};
this.updateParams = function () {
for (var m = 0; m < this.documents.length; m++) {
for (var k = 0; k < this.K; k++) {
this.thetasum[m][k] +=
(this.nd[m][k] + this.alpha) / (this.ndsum[m] + this.K * this.alpha);
}
}
for (var k = 0; k < this.K; k++) {
for (var w = 0; w < this.V; w++) {
this.phisum[k][w] +=
(this.nw[w][k] + this.beta) / (this.nwsum[k] + this.V * this.beta);
}
}
this.numstats++;
};
this.getTheta = function () {
var theta = new Array();
for (var i = 0; i < this.documents.length; i++) theta[i] = new Array();
if (this.SAMPLE_LAG > 0) {
for (var m = 0; m < this.documents.length; m++) {
for (var k = 0; k < this.K; k++) {
theta[m][k] = this.thetasum[m][k] / this.numstats;
}
}
} else {
for (var m = 0; m < this.documents.length; m++) {
for (var k = 0; k < this.K; k++) {
theta[m][k] =
(this.nd[m][k] + this.alpha) /
(this.ndsum[m] + this.K * this.alpha);
}
}
}
return theta;
};
this.getPhi = function () {
var phi = new Array();
for (var i = 0; i < this.K; i++) phi[i] = new Array();
if (this.SAMPLE_LAG > 0) {
for (var k = 0; k < this.K; k++) {
for (var w = 0; w < this.V; w++) {
phi[k][w] = this.phisum[k][w] / this.numstats;
}
}
} else {
for (var k = 0; k < this.K; k++) {
for (var w = 0; w < this.V; w++) {
phi[k][w] =
(this.nw[w][k] + this.beta) / (this.nwsum[k] + this.V * this.beta);
}
}
}
return phi;
};
})()
makeArray = (x) => {
var a = new Array();
for (var i = 0; i < x; i++) {
a[i] = 0;
}
return a;
}
make2DArray = (x, y) => {
var a = new Array();
for (var i = 0; i < x; i++) {
a[i] = new Array();
for (var j = 0; j < y; j++) a[i][j] = 0;
}
return a;
}
autoStemmer = {
const porterStemmer = await import(
"https://cdn.skypack.dev/@stdlib/nlp-porter-stemmer@0.0.7?min"
);
return porterStemmer.default;
}
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.