Determine if some text is a question

question_clf.py

import nltk

import random
from pprint import pprint

from sklearn.pipeline import Pipeline
from sklearn.metrics import classification_report
from sklearn.feature_extraction.text import HashingVectorizer

posts = nltk.corpus.nps_chat.xml_posts([
    '10-19-20s_706posts.xml',
    '11-08-20s_705posts.xml',
    '11-09-20s_706posts.xml'
])


def transform(post):
    tokens = nltk.word_tokenize(post.text)
    tagged_tokens = nltk.pos_tag(tokens)

    serialized = ['_'.join(z)
                  for z in tagged_tokens]

    text = ' '.join(serialized)

    return text, 'Question' in post.get('class')


def train_classifier():
    pipeline = Pipeline([
        ('vect', HashingVectorizer()),
        ('clf', SGDClassifier())
    ])

    featuresets = [transform(post) for post in posts]
    random.shuffle(featuresets)
    size = int(len(featuresets) * .1)
    train_set, test_set = featuresets[size:], featuresets[:size]

    X, y = zip(*train_set)
    pipeline.fit(X, y)

    X, y = zip(*test_set)
    pred = pipeline.predict(X)

    pprint([z for z in zip(X, pred, y)
            if z[1] != z[2]])

    print('accuracy %f' % pipeline.score(X, y))
    print(classification_report(y, pred))

    return pipeline


classifier = train_classifier()