#!/usr/bin/env python

import argparse
import marshal
import random
import regex

class Lipsum:
  clean = regex.compile ('\\s+$')

  """
  Create a new lipsum instance by either training a model from given input
  text file, or loading a previously saved model. Behavior actually depends
  on whether the 3rd parameter (length) is provided since it's only used when
  training a new model (but yeah, that's still an ugly hack).
  """
  def __init__ (self, path, length = None):
    # If length is none we assume input is a trained model
    if length is None:
      with open (args.model, 'rb') as file:
        (self.model, self.length) = marshal.load (file)

    # Otherwise start training
    else:
      self.length = length
      self.model = {}

      # Tokenize input string into (prefix -> (suffix, count)) tree
      tokenize = regex.compile ('((?:[-\\w]+|[^\\n.])\\s*)*(?:[\\n.]\\s*)?')
      tree = {}

      with open (path, 'r') as file:
        for match in tokenize.finditer (file.read ()):
          lexems = map (self.clean_lexem, match.captures (1))

          # Ignore empty sequences
          if len (lexems) < 1:
            continue

          # Register suffixes, including special "end of line" marker
          prefix = (None, ) * length

          for lexem in lexems + [None]:
            suffixes = tree.setdefault (prefix, {})
            suffixes[lexem] = suffixes.get (lexem, 0) + 1

            prefix = prefix[1:] + (lexem, )

      # Convert to (prefix -> (suffix, probability)) model
      for (key, suffixes) in tree.iteritems ():
        occurrences = float (sum ((count for (suffix, count) in suffixes.iteritems ())))
        thresholds = []
        total = 0

        for (lexem, count) in suffixes.iteritems ():
          total += count / occurrences

          thresholds.append ((lexem, total))

        self.model[key] = thresholds

  """
  Cleanup input lexem by squashing all "dirty" characters (see the "clean"
  regular expression above) into a single space character.
  """
  def clean_lexem (self, lexem):
    return self.clean.sub (' ', lexem).lower ()

  """
  Find first suffix above given value (used for random suffix selection).
  This method could/should be replaced by some functional call like:
  bisect (suffixes, lambda suffix: suffix[1], value)
  """
  def first_above (self, suffixes, value):
    i = 0

    while i < len (suffixes) and suffixes[i][1] <= value:
      i += 1

    return i < len (suffixes) and suffixes[i][0] or None

  """
  Generate a random lexems sequence using currently loaded model.
  """
  def generate (self):
    buffer = ''
    prefix = tuple ([None] * self.length)

    while prefix in self.model:
      lexem = self.first_above (self.model[prefix], random.random ())

      if lexem is None:
        break

      buffer += lexem
      prefix = prefix[1:] + (lexem, )

    return buffer

  """
  Save current model to file.
  """
  def save (self, path):
    with open (path, 'wb') as file:
      marshal.dump ((self.model, self.length), file)

parser = argparse.ArgumentParser (description = 'Lipsum blabla')
parser.add_argument ('-g', '--generate', type = int, default = 1, help = 'Generate N lines', metavar = 'N')
parser.add_argument ('-l', '--length', type = int, default = 3, help = 'Set prefix length', metavar = 'LEN')
parser.add_argument ('-m', '--model', action = 'store', help = 'Specify path to model', metavar = 'FILE')
parser.add_argument ('-t', '--train', action = 'store', help = 'Train from given file (and save if -m is specified)', metavar = 'FILE')

args = parser.parse_args ()

if args.train is not None:
  lipsum = Lipsum (args.train, args.length)

  if args.model is not None:
    lipsum.save (args.model)

elif args.model is not None:
  lipsum = Lipsum (args.model)

else:
  raise Exception ('please specify either model or train argument')

for i in range (args.generate):
  print lipsum.generate ()