def classify_with_expert_knowledge(paths):
sentences, labels, class_names = load_test_data(paths)
features, vocabulary, unused = extract_features_and_vocabulary(sentences)
stemmed_target_features = [stem_words(target_feature) for target_feature in TARGET_FEATURES]
predicted = classify(stemmed_target_features, features.toarray(), vocabulary)
print('EXPERT KNOWLEDGE:')
evaluate_classification(predicted, labels, sentences, class_names)
def test_extract_features(input_sentences, expected_vocabulary):
actual_features, actual_vocabulary, unused = extract_features_and_vocabulary(input_sentences)
expected_features = np.array([
[1, 1, 2, 0, 1, 1, 1, 1, 0, 0],
[0, 0, 0, 1, 0, 0, 0, 0, 1, 1]
])
assert actual_vocabulary == expected_vocabulary
assert (actual_features == expected_features).all()
def classify_with_tf_idf(paths):
sentences, labels, class_names = load_test_data(paths)
sentences = np.array(sentences)
labels = np.array(labels)
average_precisions = []
average_recalls = []
for train_index, test_index in sklearn.cross_validation.StratifiedKFold(labels, n_folds=3):
sentences_train, sentences_test = sentences[train_index], sentences[test_index]
labels_train, labels_test = labels[train_index], labels[test_index]
features_train, vocabulary, count_vectorizer = extract_features_and_vocabulary(sentences_train)
tfidf_features_train = transform_to_tfidf(features_train)
predicted = predict_with_svc(tfidf_features_train, labels_train, sentences_test, count_vectorizer)
print('TF-IDF')
average_precision, average_recall = evaluate_classification(predicted, labels_test, sentences, class_names)
average_precisions.append(average_precision)
average_recalls.append(average_recall)
evaluate_complete_classification(average_precisions, average_recalls)
def vocabulary(input_sentences):
return extract_features_and_vocabulary(input_sentences)[1]
def features(input_sentences):
return extract_features_and_vocabulary(input_sentences)[0].toarray()