text-spew.py

import argparse
import collections
import json
import sys
import random


def weighted_select(options, total):
  rmeasure = random.random()
  for option in options:
    rmeasure -= options[option]['#']/total
    if rmeasure <= 0:
      return option
  #Should never happen, but in case, retry.
  return weighted_select(options, total)


def select_char(history, model):
  tree = model['tree']
  for i in range(0, len(history)-1):
    tree = tree[history[i]]['>']
  options = tree[history[len(history)-1]]['>']
  total = tree[history[len(history)-1]]['#']
  return weighted_select(options, total)

    
def generate_sentence(model):
  history = collections.deque(['$BEGIN'])
  sentence = []
  while True:
   char = select_char(history, model)
   if char == '$END':
     break
   sentence.append(char)
   history.append(char)
   if len(history) == model['depth']:
    history.popleft()
   
  return str(''.join(sentence))



if __name__ == '__main__': 
  parser = argparse.ArgumentParser(description='Generate nonsense text from a simple language frequency model.')
  parser.add_argument('modelfile', help='The input model file built by char-generator.py')
  parser.add_argument('num_lines', help='How many lines of text to produce. Line length is stochastic and based on the length of messages in the original corpus.', type=int)
  parser.add_argument('--outfile', '-o', nargs='?', help='Writeable output file. Default is STDOUT.', default=sys.stdout, type=argparse.FileType('w'))
  args = parser.parse_args()

  model = json.load(open(args.modelfile,'r'))

  for i in range(0, args.num_lines):
    line = generate_sentence(model)
    args.outfile.write(line+'\n')