File: - Tab length: 1 2 4 8 - Lines: on off - No wrap: on off

#!/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
            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 ( ()):
                    lexems = map (self.clean_lexem, match.captures (1))

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

                    # 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:

            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: (args.model)

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

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

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