Clustring NPM | npm.io

clustring

Sort groups of strings into buckets.

API

The unit of analysis is a bucket. It looks like this:

const bucket = {
	"commonWord": 3,
	"CommonWord": 20,
	"SuperRareWord": 1
}

Binning

strcluster can break your bucket into bins by computing a bin key for each string. Here's a code sample, using the bucket from above:

import { clusterByKey } from 'clustring'
import fingerprint from 'clustring/key/fingerprint'

const clusterer = clusterByKey(bucket, fingerprint())

clusterer.cluster()
  .then(bins => { ... })
// bins is:
// [
//   {
//     "name": "CommonWord",
//     "key": "commonword",
//     "count": 23,
//     "bucket": { "commonWord": 3, "CommonWord": 20 }
//   }
// ]

KNN

strcluster can also break your bucket into bins using a distance function to compare two strings.

Distance functions aren't cheap. A block-based approach avoids comparisons by grouping strings into "blocks" that all contain the same N sequence of characters. (Effectively, this skips comparisons by assuming infinite distance if there is no such sequence). TODO: implement this

Here's some sample code:

import { clusterByKnn } from 'clustring'
import levenshtein from 'clustring/knn/levenshtein'

// levenshtein(2) is an optimization of levenshtein() that returns 0, 1, 2, or
// Infinity. You may use levenshtein(), but it's not recommended.
const clusterer = clusterByKnn(bucket, levenshtein(2), 2, { blockSize: 5 })
clusterer.cluster()
  .then(bins => { ... })
// bins will be same as in previous example, minus "key"

Progress reporting

cluster() returns a Promise immediately and processes in the background (in the current thread). It cedes control to the event loop every few milliseconds so your app remains responsive.

To track progress, try something like this:

const clusterer = clusterByKey(bucket, fingerprint(), { tickMs: 8 })

let timeout = null
function reportProgressAndReschedule () {
  console.log('Progress: ', clusterer.progress)
  timeout = setTimeout(reportProgressAndReschedule, 1)
}
// start progress-report loop
timeout = setTimeout(reportProgressAndReschedule, 1)

clusterer.cluster()
  .then(bins => {
    clearTimeout(timeout)
    // ... handle bins
  })

During cluster(), clustring will periodically check whether it has blocked the main thread for more than tickMs milliseconds. if it has, it will cede control to the event loop for one event-loop "tick" before resuming. Your setTimeout() callback will only be called once cluster() cedes control, even though it requests to be called after 1 millisecond.

Cancellation

If you wish to stop clustering, run clusterer.cancel(). Of course, you can only execute clusterer.cancel() during a tick, so consider your tickMs.

Developing

npm install
npm test -- --watch             # runs tests continuously
npm run-script build -- --watch # builds continuously

Pick a feature; write a test; make it pass; commit.

Deploying

Update version in package.json
npm install to update package-lock.json
git commit -am 'vx.x.x && git tag vx.x.x && git push && git push origin vx.x.x
npm publish