Skip to content

spencermountain/efrt

Repository files navigation

compression of key-value data
npm install efrt

if your data looks like this:

var data = {
  bedfordshire: 'England',
  aberdeenshire: 'Scotland',
  buckinghamshire: 'England',
  argyllshire: 'Scotland',
  bambridgeshire: 'England',
  cheshire: 'England',
  ayrshire: 'Scotland',
  banffshire: 'Scotland'
}

you can compress it like this:

import { pack } from 'efrt'
var str = pack(data)
//'England:b0che1;ambridge0edford0uckingham0;shire|Scotland:a0banff1;berdeen0rgyll0yr0;shire'

then _very!_ quickly flip it back into:

import { unpack } from 'efrt'
var obj = unpack(str)
obj['bedfordshire'] //'England'

Yep,

efrt packs category-type data into a very compressed prefix trie format, so that redundancies in the data are shared, and nothing is repeated.

By doing this clever-stuff ahead-of-time, efrt lets you ship much more data to the client-side, without hassle or overhead.

The whole library is 8kb, the unpack half is barely 2kb.

it is based on:

Benchmarks!

Basically,
  • get a js object into very compact form
  • reduce filesize/bandwidth a bunch
  • ensure the unpacking time is negligible
  • keep word-lookups on critical-path
import { pack, unpack } from 'efrt' // const {pack, unpack} = require('efrt')

var foods = {
  strawberry: 'fruit',
  blueberry: 'fruit',
  blackberry: 'fruit',
  tomato: ['fruit', 'vegetable'],
  cucumber: 'vegetable',
  pepper: 'vegetable'
}
var str = pack(foods)
//'{"fruit":"bl0straw1tomato;ack0ue0;berry","vegetable":"cucumb0pepp0tomato;er"}'

var obj = unpack(str)
console.log(obj.tomato)
//['fruit', 'vegetable']

or, an Array:

if you pass it an array of strings, it just creates an object with true values:

const data = [
  'january',
  'february',
  'april',
  'june',
  'july',
  'august',
  'september',
  'october',
  'november',
  'december'
]
const packd = pack(data)
// true¦a6dec4febr3j1ma0nov4octo5sept4;rch,y;an1u0;ly,ne;uary;em0;ber;pril,ugust
const sameArray = Object.keys(unpack(packd))
// same thing !

Reserved characters

the keys of the object are normalized. Spaces/unicode are good, but numbers, case-sensitivity, and some punctuation (semicolon, comma, exclamation-mark) are not (yet) supported.

specialChars = new RegExp('[0-9A-Z,;!:|¦]')

efrt is built-for, and used heavily in compromise, to expand the amount of data it can ship onto the client-side. If you find another use for efrt, please drop us a line🎈

Performance

efrt is tuned to be very quick to unzip. It is O(1) to lookup. Packing-up the data is the slowest part, which is usually fine:

var compressed = pack(skateboarders) //1k words (on a macbook)
var trie = unpack(compressed)
// unpacking-step: 5.1ms

trie.hasOwnProperty('tony hawk')
// cached-lookup: 0.02ms

Size

efrt will pack filesize down as much as possible, depending upon the redundancy of the prefixes/suffixes in the words, and the size of the list.

  • list of countries - 1.5k -> 0.8k (46% compressed)
  • all adverbs in wordnet - 58k -> 24k (58% compressed)
  • all adjectives in wordnet - 265k -> 99k (62% compressed)
  • all nouns in wordnet - 1,775k -> 692k (61% compressed)

but there are some things to consider:

  • bigger files compress further (see 🎈 birthday problem)
  • using efrt will reduce gains from gzip compression, which most webservers quietly use
  • english is more suffix-redundant than prefix-redundant, so non-english words may benefit from other styles

Assuming your data has a low category-to-data ratio, you will hit-breakeven with at about 250 keys. If your data is in the thousands, you can very be confident about saving your users some considerable bandwidth.

Use

IE9+

<script src="https://unpkg.com/efrt@latest/builds/efrt.min.cjs"></script>
<script>
  var smaller = efrt.pack(['larry', 'curly', 'moe'])
  var trie = efrt.unpack(smaller)
  console.log(trie['moe'])
</script>

if you're doing the second step in the client, you can load just the CJS unpack-half of the library(~3k):

const unpack = require('efrt/unpack') // node/cjs
<script src="https://unpkg.com/efrt@latest/builds/efrt-unpack.min.cjs"></script>
<script>
  var trie = unpack(compressedStuff)
  trie.hasOwnProperty('miles davis')
</script>

Thanks to John Resig for his fun trie-compression post on his blog, and Wiktor Jakubczyc for his performance analysis work

MIT

About

neato compression for key-value data

Topics

Resources

License

Stars

Watchers

Forks