def train(self):
# Maybe some stemming later?
docCount = {1:utility.Counter(), 2:utility.Counter()}
for text in self.documents.itervalues():
docCount[1].update(set(utility.ngramFinder(text, 1)))
docCount[2].update(set(utility.ngramFinder(text, 2)))
print "Got doc count"
categoryCount = utility.Counter()
categoryWords = defaultdict(list)
for q in Question.objects.all()[::100]:
words = utility.wordParse(q.body)
categoryWords[q.category] += words
categoryWords[""] += words
categoryBins = len(set(categoryWords[""]))
del categoryWords[""]
for category in categoryWords:
categoryCount.update((word,) for word in set(categoryWords[category]))
for category, words in categoryWords.items():
categoryWords[category] = WordDist(words, categoryBins)
utility.wordFilter(categoryCount, len(categoryWords),
categoryWords[category])
print "Trained Category"
for label in self.documents:
self.features[label] = NGramModel(2, self.documents[label],
docCount, len(self.documents))
category = Label.objects.get(body=label).questions.all()[0].category
self.features[label].addBackoff(categoryWords[category],
categoryBins)
print "Trained Wikipedia"
return categoryCount
def __init__(self, n, text, docCount, numDoc):
self.n = n
self.model = []
for i in xrange(1, n + 1):
ngrams = utility.ngramFinder(text, i)
cwd = defaultdict(functools.partial(WordDist, None, len(docCount[i])))
for ngram in ngrams:
context = tuple(ngram[:-1])
token = ngram[-1]
cwd[context].add(token)
for context in cwd.keys():
utility.wordFilter(docCount[i], numDoc, cwd[context], context)
self.model.append(cwd)