Generate random words

README.md

Generating weights

Before generating any words, you first need a weights file. Generate one using build_weights.rb like so:

cat /path/to/dictionary | ./build_weights.rb 5 > weights.json

This will generate the weights.json file based on the words in /path/to/dictionary with a look-back of 5. The input dictionary is a collection of words, one per line. The look-back controls how much context will be used when generating a word. A look-back of one tends to generate more gibberish words, higher values will create more coherent words. 5 seems to give very good results.

Generating words

Once you have a weight file, you can generate words using make_word.rb like so:

./make_word.sh weights.json

build_weights.rb

#!/usr/bin/env ruby

require 'json'

LOOK_BACK = ARGV[0].to_i

counts = STDIN.readlines.map(&:downcase).uniq
  .map(&:chars).flat_map do |chars|
    last_state = chars.first

    chars.slice(1..chars.length).map do |ch|
      ch = '' if ch == "\n"

      tr = [last_state, ch]

      last_state = (last_state + ch).chars.last(LOOK_BACK).join

      tr
    end
  end.reduce({}) do |memo, (start, result)|
    memo[start] ||= {}
    memo[start][result] ||= 0
    memo[start][result] += 1

    memo
  end

weights = counts.merge(counts) do |_, count|
  total = count.values.inject(:+)
  acc = 0

  count.merge(count) do |_, value|
    acc += Rational(value, total).to_f
  end
end

puts(JSON.dump(weights))

make_word.rb

#!/usr/bin/env ruby

require 'json'

weights_file = File.join(File.dirname(__FILE__), ARGV[0])
weights = JSON.parse(File.read(weights_file))

LOOK_BACK = weights.keys.map(&:length).max

state = 'abcdefghijklmnopqrstuvwxyz'.chars.sample
result = []

loop do
  result << state

  r = rand
  state = weights[result.last(LOOK_BACK).join].find { |ch, prob| prob > r }[0]

  break if state == ''
end

puts(result.join)