LDA-ts

LDA-ts is a TypeScript implementation of Topic modeling (LDA: Latent Dirichlet Allocation) using collapsed gibbs sampling.

Installation

From GitHub Packages

npm install @y-ota/lda-ts

or

From this repository

npm install git+https://github.com/y-ota/LDA-ts.git

Usage

import ldats from "@y-ota/lda-ts";

const X = [[1,0,0,1,0,1],[0,1,1,1,0,1],[1,1,1,1,0,1],...] // ex: Bag of words
const lda = new ldats.LDA();
const model = lda.fit(X);
console.log(model);

Output of console.log(model);

 {
     D: 4,
     V: 3,
     K: 10,
     alpha: 0.5,
     beta: 0.1,
     X: [ [ 1, 0, 1 ], [ 1, 0, 1 ], [ 1, 0, 1 ], [ 1, 0, 1 ] ],
     iters: 10,
    ...
     phi: [
      [ 0.04347826086956522, 0.9130434782608696, 0.04347826086956522 ],
      [ 0.9130434782608696, 0.04347826086956522, 0.04347826086956522 ],
      [ 0.33333333333333337, 0.6363636363636365, 0.030303030303030307 ],
      …
     ],
     theta: [
       [
         0.25,
         0.17857142857142858,
         0.10714285714285714,
         0.10714285714285714,
         0.03571428571428571,
         0.03571428571428571,
         0.10714285714285714,
         0.03571428571428571,
         0.10714285714285714,
         0.03571428571428571
       ],
       [
         0.4473684210526316,
         0.02631578947368421,
         0.02631578947368421,
...
         

API documentation

LDA

constructor

Arguments

Property	Explanation	Default value
K	Number of topics	10
alpha	Hyper parameteter Alpha	0.5
beta	Hyper parameteter Beta	0.1
nIter	Number of the Gibbs Sampling iterations	10

fit

Arguments

Property	Explanation
X	All documents (ex: Bang of Words)

Returns

Property	Explanation
Model	Estimated model

Model

Property	Explanation
X	All documents
D	Number of documents
V	Number of vocabulary
K	Number of topics
alpha	Hyper parameter Alpha
beta	Hyper parameter Beta
nIter	Number of the Gibbs Sampling iterations
theta	Estimated topic distributions
phi	Estimated word distributions
z	Assigned topics for word (D * V)
nkw	Number of words assinged to topic (V * K)
ndk	Number of words assigned to topic in document (D * K)
nk	Number of topics in all documents (1 * K)
nd	Number of worts in document (1 * D)

License

© 2020 Yusuke Ota, LDA-ts released under the MIT Lincense