MagicCloud - simple pretty word cloud for Ruby

MagicCloud is simple, pure-ruby library for making pretty Wordle-like clouds. It uses RMagick as graphic backend.

Usage

words = [
  ['test', 50],
  ['me', 40],
  ['tenderly', 30],
  # ....
]
cloud = MagicCloud::Cloud.new(words, rotate: :free, scale: :log)

# To save to file, if not redering it for a page
img = cloud.draw(960, 600) #default height/width
img.write('test.png')

Or from command-line:

./bin/magic_cloud --textfile samples/cat-in-the-hat.txt -f test.png --rotate free --scale log

Resulting in:

Installation

gem install magic_cloud

rmagick is requirement, and it needs compilation, so you may expect problems in non-compiler-friendly environment (Windows).

Origins

At first, it was a straightforward port of d3.layout.cloud.js by Jason Davies, which, I assume, is an implementation of Wordle algorithm.

Then there was major refactoring, to make code correspond to Ruby standards (and be understandable to poor dumb me).

Then collision algorithm was rewritten from scratch.

And now we are here.

References:

• https://github.com/jasondavies/d3-cloud
• http://stackoverflow.com/questions/342687/algorithm-to-implement-a-word-cloud-like-wordle
• http://static.mrfeinberg.com/bv_ch03.pdf

Performance

It's reasonable for me. On my small Thinkpad E330, some 50-words cloud image, size 700×500, are typically generated in <3sec. It's not that cool, yet not too long for you to fell asleep.

The time of cloud making depends on words count, size of image (it's faster to find place for all words on larger image) and used rotation algorithm (vertical+horizontal words only is significantly faster - and, on my opinion, better looking - than "cool" free-rotated-words cloud). It even depends on font - dense font like Impact takes more time to lay out than sparse Tahoma.

Major performance eater is perfect collision detection, which Wordle-like cloud needs. MagicCloud for now uses really dumb algorithm with some not-so-dumb optimizations. You can look into lib/magic_cloud/collision_board.rb - everything can be optimized is there; especially in CollisionBoard#collides? method.

I assume, for example, that naive rewriting of code in there as a C extension would help significantly.

Another possible way is adding some smart tricks, which eliminate as much of pixel-by-pixel comparisons as possible (some of already made are criss-cross intersection check, and memoizing of last crossed sprite).

Memory effectiviness

Basically: it's not.

Plain Ruby arrays are used to represent collision bitmasks (each array member stand for 1 bit), so, for example, 700×500 pixel cloud will requre collision board size 700*500 (i.e. 350k array items only for board, and slightly less for all sprites).

It should be wise to use some packing (considering each Ruby Fixmnum can represent not one, but whole 32 bits). Unfortunately, all bit array libraries I've tried are causing major slowdown of cloud computation. With, say, 50 words we'll have literally millions of operation bitmask#[] and bitmask#[]=, so, even methods like Fixnum#& and Fixnum#| (typically used for bit array representation) are causing significant overload.

Configuration

cloud = MagicCloud.new(words, palette: palette, rotate: rotate)

• :palette (default is :color20):
  • :category10, :category20, ... - from D3.js
  • [array, of, colors] - each color should be hex color, or any other RMagick color string (See "Color names at http://www.imagemagick.org/RMagick/doc/imusage.html)
  • any lambda, accepting (word, index) and returning color string
  • any object, responding to color(word, index) - so, you can make color depend on tag text, not only on its number in tags list
• :rotate - rotation algorithm:
  • :square (only horizontal and vertical words) - it's default
  • :none - all words are horizontal (looks boooring)
  • :free - any word rotation angle, looks cool, but not very readable and slower to layout
  • [array, of, angles] - each of possible angle should be number 0..360
  • any lambda, accepting (word, index) and returning 0..360
  • any object, responding to rotate(word, index) and returning 0..360
• :scale - how word sizes would be scaled to fit into (FONT_MIN..FONT_MAX) range:
  • :no - no scaling, all word sizes are treated as is;
  • :linear - linear scaling (default);
  • :log - logarithmic scaling;
  • :sqrt - square root scaling;
• :font_family (Impact is default, ex. Arial, Helvetica, Futura).
• :font - Full path to custom font file. Overwrite :font_family.

Current state

This library is extracted from a real-life project. It should be pretty stable (apart from bugs introduced during extraction and gemification).

What it really lacks for now, is thorough (or any) testing, and some more configuration options.

Also, while core algorithms (collision_board.rb, layouter.rb) are pretty accurately written and documented, "wrapping code" (options parsing and so on) are a bit more chaotic - it's subject to refactor and cleanup.

All feedback, usage examples, bug reports and feature requests are appreciated!