node-crfsuite

A nodejs binding for crfsuite

This is a link to the CRFSuite library written by Naoaki Okazaki. CRF or Conditional Random Fields are a class of statistical modeling method often applied in pattern recognition and machine learning and used for structured prediction.

Installation

For most "standard" use cases (on Mac, Linux, or Windows on a x86 or x64 processor), node-crfsuite will install easy with:

npm install crfsuite

API Usage

CRFSuite Tagger

const crfsuite = require('crfsuite')
const tagger = new crfsuite.Tagger()

let is_opened = tagger.open('./path/to/crf.model')
console.log('File model is opened:', is_opened)

let tags = tagger.tag(input)
console.log('Tags: ', tags)

CRFSuite Trainer

const path = require('path')
const crfsuite = require('crfsuite')
const trainer = new crfsuite.Trainer({
  debug: true
})

let model_filename = path.resolve('./model.crfsuite')

let xseq = [['walk'], ['walk', 'shop'], ['clean', 'shop']]
let yseq = ['sunny', 'sunny', 'rainy']

// submit training data to the trainer
trainer.append(xseq, yseq)
trainer.train(model_filename)

// output: ./model.crfsuite

Installation Special Cases

We use node-pre-gyp to compile and publish binaries of the library for most common use cases (Linux, Mac, Windows on standard processor platforms). If you have a special case, node-crfsuite will work, but it will compile the binary during the install. Compiling with nodejs is done via node-gyp which requires Python 2.x, so please ensure you have it installed and in your path for all operating systems. Python 3.x will not work.

• See node-gyp installation prerequisites.

Build from source

# clone the project
git clone --recursive https://github.com/vunb/node-crfsuite.git

# go to working folder
cd node-crfsuite

# install dependencies and build the binary
npm install

For development:

# rebuild
npm run build

# run unit-test
npm test

Change Log

2019-07-18: Release version 1.0.0

• Convert all sources to use N-API, remove nan
• Add typescript declaration file
• Cleanup package dependencies
• CI Tool run & test only in node version 10 LTS
• Add project convention: .gitattributes, .editorconfig
• Enforced to use new keyword to create new Tagger and Trainer

2019-06-09: uses N-API

From [email protected] the library uses N-API to use the binary in multiple version of Node.

• Acording to Node.js Foundation: With N-API, native module developers can compile their module once per platform and architecture, and make it available for any version of Node.js that implements N-API. This holds true even for versions of Node.js that are built with a different VM, e.g. Node-ChakraCore.

Contributing

Pull requests and stars are highly welcome.

For bugs and feature requests, please create an issue.