derhuerst / nbayes

A Naive Bayes classifier written in JavaScript.

nbayes

nbayes is a lightweight Naive Bayes Classifier written in vanilla JavaScript. It classifies a document (arbitrary piece of text) among the classes (arbitrarily named categories) it has been trained with before. This is all based on simple mathematics. As an example, you could use nbayes to answer the following questions.

• Is an email spam, or not spam ?
• Is a news article about technology, politics, or sports ?
• Does a piece of text express positive emotions, or negative emotions?

const nbayes = require('nbayes')

let classifier = nbayes()
classifier.learn('happy',   nbayes.stringToDoc('amazing, awesome movie!! Yeah!! Oh boy.'))
classifier.learn('happy',   nbayes.stringToDoc('Sweet, this is incredibly amazing, perfect, great!!'))
classifier.learn('angry',   nbayes.stringToDoc('terrible, shitty thing. Damn. This Sucks!!'))
classifier.learn('neutral', nbayes.stringToDoc('I dont really know what to make of this.'))

classifier.classify(nbayes.stringToDoc('awesome, cool, amazing!! Yay.'))
// -> 'happy'

nbayes offers a simple and straightforward API, keeping it below 3kb (minified). It is a rewrite of ttezel/bayes and thoroughly tested.

Installing

npm install nbayes

API

nbayes.createDoc()

Creates a representation of a document, which can be used to track words and their frequencies.

Example

let d = nbayes.createDoc()
d.set('foo', 2)
d.add('bar')
d.increase('bar', 2)

d.has('FOO') // -> false
d.get('foo') // -> 2
d.get('bar') // -> 3
d.sum() // -> 5
d.words() // -> ['foo', 'bar']

Methods

• has(word): If word has been added before.
• get(word): Returns the count of word.
• set(word, count): Sets the count of word.
• add(word): Shorthand for increase(word, 1).
• increase(word, d = 1): Adds d to the count of word.
• sum: Returns the sum of all word counts.
• words: Returns the distinct words.

nbayes.stringToDoc()

Returns a document from the string. Special characters will be ignored. Everything will be lowercase.

Note: It is probably a better idea to use a proper tokenizer/stemmer and to remove stopwords to support non-Latin languages and to get more accurate results.

nbayes.stringToDoc('awesome, amazing!! Yay.').words()
// -> ['awesome', 'amazing', 'yay']

nbayes()

Creates a classifier, which can learn and then classify documents into classes.

Example

let c = nbayes()
c.learn('happy',   nbayes.stringToDoc('amazing, awesome movie!! Yeah!! Oh boy.'))
c.learn('happy',   nbayes.stringToDoc('Sweet, this is incredibly amazing, perfect, great!!'))
c.learn('angry',   nbayes.stringToDoc('terrible, shitty thing. Damn. This Sucks!!'))
c.learn('neutral', nbayes.stringToDoc('I dont really know what to make of this.'))

c.classify(nbayes.stringToDoc('awesome, cool, amazing!! Yay.'))
// -> 'happy'
c.probabilities(nbayes.stringToDoc('awesome, cool, amazing!! Yay.'))
// -> { happy: 0.000001…,
//      angry: 2.384…e-7,
//      neutral: 1.665…e-7 }

Methods

• learn(class, doc): Tags words of doc as being of class.
• probabilities(doc): For each stored class, returns the probability of doc, given the class.
• classify(doc): For doc, returns the class with the highest probability.
• prior(class): Computes the probability of class out of all classes.
• likelihood(class, doc): Computes the probability of doc, given class.

Contributing

If you have a question, found a bug or want to propose a feature, have a look at the issues page.

nbayes's Issues

Enhance README.md with badges

![latest release](https://img.shields.io/github/release/derhuerst/nbayes.svg)
![npm version](https://img.shields.io/npm/v/nbayes.svg)
![node version](https://img.shields.io/node/v/nabyes.svg)
![bower version](https://img.shields.io/bower/v/nabyes.svg)
![codeclimate coverage](https://img.shields.io/codeclimate/coverage/github/derhuerst/nbayes.svg)

Write documentation.

Nothing more to explain.

Write unit tests and use Travis CI

Also cover all private methods and Vocabulary and BagOfWords.

Link-Rot in README

There are two links in the README which point to a now-deleted video on YouTube (i.e. youtube.com/watch?v=DdYSMwEWbd4). While the video is no longer available, material by one of the authors is available here: https://web.stanford.edu/~jurafsky/slp3/4.pdf. Presumably the content is similar, if not the same.

Should the link be updated to point instead to the aforementioned PDF, or perhaps to another video (e.g. https://www.youtube.com/watch?v=rmVRLeJRkl4)?