#!/usr/local/bin/python

import sys, math, random

import optparse, fileinput
optparser = optparse.OptionParser(usage="%prog [options]")
optparser.add_option('-m', '--model', dest='model_filename', type='str', default=None, help='language model file')
optparser.add_option('-o', '--output', dest='output_filename', type='str', default="-", help='output file')
optparser.add_option('-r', '--random', dest='n_random', type='int', default="1", help='number of random sentences to generate')
(opts,args) = optparser.parse_args()

if opts.model_filename is None:
    sys.stderr.write("Error: must provide exactly one language model file\n")
    sys.exit(1)
model_file = file(opts.model_filename)

ngrams = {}
n = None
for line in model_file:
    (ngram,p) = line.split("\t")
    ngram = tuple(ngram.split())
    if n is None:
        n = len(ngram)
    ngrams.setdefault(ngram[:-1], {}).setdefault(ngram[-1], float(p))

if opts.output_filename == "-":
    output_file = sys.stdout
else:
    output_file = file(opts.output_filename, "w")

for i in xrange(opts.n_random):
    context = ("<s>",)*(n-1)
    output = []
    for j in xrange(1000): # max sentence length
        pc = 0.0
        r = random.random()
        for (word, p) in ngrams.get(context,{}).iteritems():
            pc += p
            if pc >= r:
                if word != "</s>":
                    output.append(word)
                    context = (context+(word,))[-n+1:]
                break
        else:
            # this shouldn't happen
            output.append("?")
        if word == "</s>":
            break
    output_file.write(" ".join(output) + "\n")