class Glados(object):
def __init__(self, data_filename):
self.logger = logging.getLogger("Glados")
self.file = File()
self.parser = SentenceParser()
self.data_filename = data_filename
self.classifier = self.train_and_get_classifier(self.data_filename)
"""
Public api
input: user question text
output: answer
"""
def get_help(self, question):
features = self.parser.extract_feature(question)
print('features', features)
answer = self.classifier.classify(features)
prob = self.classifier.prob_classify(features).prob(answer)
# self.logger.info('features for question are %s', features)
print('Answer:', answer, "(", prob, ")")
response = dict(question=question, answer=answer, probability=prob)
return response
def train_and_get_classifier(self, data_filename):
split_ratio = 0.8
data = self.parser.get_content(data_filename)
data_set = self.parser.extract_feature_from_doc(data)
random.shuffle(data_set)
data_length = len(data)
train_split = int(data_length * split_ratio)
training_data = data_set[:train_split]
self.error_analysis_for_text_input_to_feature_extraction(data)
self.error_analysis_for_features_to_predicted_answer(training_data)
test_data = data_set[train_split:]
# self.logger.debug('\n'.join([str(x) for x in data_set]))
classifier, classifier_name, test_set_accuracy, training_set_accuracy = self.train_using_naive_bayes(
training_data, test_data)
self.file.append(
get_module_path("output/accuracy.txt"),
"\n%s\t\t%s\t\t\t%.8f\t\t%.8f" %
(classifier_name, data_length, training_set_accuracy,
test_set_accuracy))
return classifier
def error_analysis_for_text_input_to_feature_extraction(self, data):
text_features_answer = []
for (text, category, answer) in data:
sent_features = self.parser.extract_feature(text)
text_features_answer.append((text, sent_features, answer))
# logging.debug(datetime.datetime.now())
# print('>>>>>>>>TEXT>>>>>>>>>FEATURES>>>>>>ANSWER:\n ', str(json.dumps(text_features_answer)))
def error_analysis_for_features_to_predicted_answer(self, test_data):
classifier = nltk.NaiveBayesClassifier.train(test_data)
classifier.show_most_informative_features()
errors = []
for (feature, actual_output) in test_data:
guess = classifier.classify(feature)
if (guess != actual_output):
errors.append((feature, actual_output, guess))
print(
'>>>>>FEATURE>>>>>>>>>ACUTUAL OUTPUT><<<<<<<<PREDICTION<<<<<<<<<<<<<<<'
)
print('Errors:', json.dumps(errors))
def train_using_naive_bayes(self, training_data, test_data):
classifier = nltk.NaiveBayesClassifier.train(training_data)
classifier_name = type(classifier).__name__
training_set_accuracy = nltk.classify.accuracy(classifier,
training_data)
test_set_accuracy = nltk.classify.accuracy(classifier, test_data)
return classifier, classifier_name, test_set_accuracy, training_set_accuracy
