def baseline(tweets_train, train_labels, tweets_test, test_labels):
# Import the subjectivity lexicon
subj_dict = data_proc.get_subj_lexicon()
types_of_features = ['1', '2', '3', 'ngrams']
for t in types_of_features:
start = time.time()
utils.print_model_title("Classification using feature type " + t)
if t is '1':
x_train_features = extract_baseline_features.get_features1(
tweets_train, subj_dict)
x_test_features = extract_baseline_features.get_features1(
tweets_test, subj_dict)
if t is '2':
x_train_features = extract_baseline_features.get_features2(
tweets_train, subj_dict)
x_test_features = extract_baseline_features.get_features2(
tweets_test, subj_dict)
if t is '3':
x_train_features = extract_baseline_features.get_features3(
tweets_train, subj_dict)
x_test_features = extract_baseline_features.get_features3(
tweets_test, subj_dict)
if t is 'ngrams':
ngram_map, x_train_features = extract_baseline_features.get_ngram_features(
tweets_train, n=1)
x_test_features = extract_baseline_features.get_ngram_features_from_map(
tweets_test, ngram_map, n=1)
# Get the class ratio
class_ratio = utils.get_classes_ratio_as_dict(train_labels)
# Train on a Linear Support Vector Classifier
print("\nEvaluating a linear SVM model...")
classifiers.linear_svm(x_train_features, train_labels, x_test_features,
test_labels, class_ratio)
# Train on a Logistic Regression Classifier
print("\nEvaluating a logistic regression model...")
classifiers.logistic_regression(x_train_features, train_labels,
x_test_features, test_labels,
class_ratio)
end = time.time()
print(
"Completion time of the baseline model with features type %s: %.3f s = %.3f min"
% (t, (end - start), (end - start) / 60.0))
def run_classifiers_with_doc2vec(reviews, scores, lang='en'):
'''Corpus should be an array of TaggedDocument objects.'''
corpus = list(get_corpus(reviews, scores, lang))[:20000]
train_corpus, test_corpus = train_test_split(corpus, test_size=0.25, random_state=42)
doc2vec_model = create_doc2vec_model(train_corpus)
train_targets, train_regressors = zip(*[(doc.words, doc.tags[0]) for doc in train_corpus])
test_targets, test_regressors = zip(*[(doc.words, doc.tags[0]) for doc in test_corpus])
'''
For every review, we apply doc2vec_model.infer_vector(review). This creates
a feature vector for every document (in our case, review) in the corpus.
'''
train_x, train_y = get_train_lists(doc2vec_model, train_targets, train_regressors)
test_x, test_y = get_test_lists(doc2vec_model, test_targets, test_regressors)
classifier = classifiers.logistic_regression(train_x, train_y)
# classifier = classifier_func(train_x, train_y)
# return logistic_reg
train_predictions = classifier.predict(train_x) # Training
train_accuracy = metrics.accuracy_score(train_y, train_predictions)
class_probabilities_train = classifier.predict_proba(train_x)
train_auc_score = metrics.roc_auc_score(train_y, class_probabilities_train[:, 1]);
print('\nTraining:')
print(' accuracy:', format(100 * train_accuracy, '.2f'))
print(' AUC value:', format(100 * train_auc_score, '.2f'))
test_predictions = classifier.predict(test_x) # Test
test_accuracy = metrics.accuracy_score(test_y, test_predictions)
class_probabilities_test = classifier.predict_proba(test_x)
test_auc_score = metrics.roc_auc_score(test_y, class_probabilities_test[:, 1]);
print('\nTesting:')
print(' accuracy:', format(100 * test_accuracy, '.2f'))
print(' AUC value:', format(100 * test_auc_score, '.2f'))
def perform_cross_validation_split(data, num_splits, filename):
y = data['defect_status']
del data["defect_status"] # removing defect answer for training
X = data.to_numpy() # remove defects
# performing cross-validation
kf = KFold(n_splits=num_splits)
for train, test in kf.split(X):
X_train, X_test = X[train], X[test]
y_train, y_test = y[train], y[test]
actual = y_test.tolist()
# Classification and Regression Tree (CART)
prediction = classifiers.cart(X_train, y_train, X_test)
calculate_metrics(prediction, actual, "cart")
# KNearestNeighbor(KNN)
prediction = classifiers.knn(X_train, y_train, X_test)
calculate_metrics(prediction, actual, "knn")
# Logistic Regression (LR)
prediction = classifiers.logistic_regression(X_train, y_train, X_test)
calculate_metrics(prediction, actual, "lr")
# Naive Bayes (NB)
prediction = classifiers.naive_bayes(X_train, y_train, X_test)
calculate_metrics(prediction, actual, "nb")
# Random Forest (RF)
prediction = classifiers.random_forest(X_train, y_train, X_test)
calculate_metrics(prediction, actual, "rf")
print("\nCalculate mean metrics...")
filename = filename.split("IST_")[1].split(".csv")[0]
precision, recall, f_score, auc = calculate_mean_metrics(
precision_results_cart, recall_results_cart, f_score_results_cart,
auc_results_cart)
save_results(precision, recall, f_score, auc, filename, "CART")
precision, recall, f_score, auc = calculate_mean_metrics(
precision_results_knn, recall_results_knn, f_score_results_knn,
auc_results_knn)
save_results(precision, recall, f_score, auc, filename, "KNN")
precision, recall, f_score, auc = calculate_mean_metrics(
precision_results_lr, recall_results_lr, f_score_results_lr,
auc_results_lr)
save_results(precision, recall, f_score, auc, filename, "LR")
precision, recall, f_score, auc = calculate_mean_metrics(
precision_results_nb, recall_results_nb, f_score_results_nb,
auc_results_nb)
save_results(precision, recall, f_score, auc, filename, "NB")
precision, recall, f_score, auc = calculate_mean_metrics(
precision_results_rf, recall_results_rf, f_score_results_rf,
auc_results_rf)
save_results(precision, recall, f_score, auc, filename, "RF")
def baseline(tweets_train, train_labels, tweets_test, test_labels):
subj_dict = dproc.get_subj_lexicon('hindi_lexicon.tff')
types_of_features = ['1', '2', 'ngrams'] # '3' is removed
for t in types_of_features:
start = time.time()
utils.print_model_title("Classification using features type " + t)
if t is '1':
x_train_features = extract_baseline_features.get_features1(
tweets_train, subj_dict)
x_test_features = extract_baseline_features.get_features1(
tweets_test, subj_dict)
if t is '2':
x_train_features = extract_baseline_features.get_features2(
tweets_train, subj_dict)
x_test_features = extract_baseline_features.get_features2(
tweets_test, subj_dict)
#if t is '3':
# x_train_features = extract_baseline_features.get_feature3(tweets_train, subj_dict)
# x_test_features = extract_baseline_features.get_feature3(tweets_test, subj_dict)
if t is 'ngrams':
ngram_map, x_train_features = extract_baseline_features.get_ngram_features(
tweets_train, n=1)
x_test_features = extract_baseline_features.get_ngram_features_from_map(
tweets_test, ngram_map, n=1)
#get the class ratio
class_ratio = utils.get_classes_ratio_as_dict(train_labels)
# train on a linear Support Vector Classifer
print('\n Evaluating a linear SVM model...')
classifiers.linear_svm(x_train_features, train_labels, x_test_features,
test_labels, class_ratio)
#train on logistic regression
classifiers.logistic_regression(x_train_features, train_labels,
x_test_features, test_labels,
class_ratio)
end = time.time()
print(
"Completion time of the baseline model with features type %s: %.3f s = %.3f min"
% (t, (end - start), (end - start) / 60.0))
def supperclassify(train_set, train_label, test_set, test_label):
'''Different methods'''
train_voted = voting(train_set)
aux = train_voted == train_label
correct = sum(aux.astype(int))
_accuracy = (correct * 100) / len(train_label)
_precision, _recall, _f1score, _support = ut.get_measures_for_each_class(
train_label, train_voted)
print 'Estimator VOTING'
print 'Average Accuracy:\t', _accuracy
print 'Average Precision:\t', _precision
print 'Average Recall:\t', _recall
print 'Average F1 Measure:\t', _f1score
print '\n'
lambdas = weighted_voting_getlambdas(train_set, train_label)
results = weighted_voting(test_set, lambdas)
aux = results == test_label
correct = sum(aux.astype(int))
_accuracy = (correct * 100) / len(test_label)
_precision, _recall, _f1score, _support = ut.get_measures_for_each_class(
test_label, results)
print 'Estimator W_VOTING'
print 'Average Accuracy:\t', _accuracy
print 'Average Precision:\t', _precision
print 'Average Recall:\t', _recall
print 'Average F1 Measure:\t', _f1score
rf = clf.classifier_randomForest(train_set, train_label)
results = clf.evaluateResults(rf,
test_set,
test_label,
estimator_name='RF')
lr = clf.logistic_regression(train_set, train_label)
results = clf.evaluateResults(lr,
test_set,
test_label,
estimator_name='LR')
svm = clf.classifier_svm(train_set, train_label)
results = clf.evaluateResults(svm,
test_set,
test_label,
estimator_name='SVM')
rbf = clf.rbf_classifier(train_set, train_label)
results = clf.evaluateResults(rbf,
test_set,
test_label,
estimator_name='RBF')
def supperclassify(train_set, train_label, test_set, test_label):
'''Different methods'''
train_voted = voting(train_set)
aux = train_voted == train_label
correct = sum(aux.astype(int))
_accuracy = (correct * 100) / len(train_label)
_precision, _recall, _f1score, _support = ut.get_measures_for_each_class(train_label, train_voted)
print 'Estimator VOTING'
print 'Average Accuracy:\t', _accuracy
print 'Average Precision:\t', _precision
print 'Average Recall:\t', _recall
print 'Average F1 Measure:\t', _f1score
print '\n'
lambdas = weighted_voting_getlambdas(train_set, train_label)
results = weighted_voting(test_set, lambdas)
aux = results == test_label
correct = sum(aux.astype(int))
_accuracy = (correct * 100) / len(test_label)
_precision, _recall, _f1score, _support = ut.get_measures_for_each_class(test_label, results)
print 'Estimator W_VOTING'
print 'Average Accuracy:\t', _accuracy
print 'Average Precision:\t', _precision
print 'Average Recall:\t', _recall
print 'Average F1 Measure:\t', _f1score
rf = clf.classifier_randomForest(train_set, train_label)
results = clf.evaluateResults(rf, test_set, test_label, estimator_name='RF')
lr = clf.logistic_regression(train_set, train_label)
results = clf.evaluateResults(lr, test_set, test_label, estimator_name='LR')
svm = clf.classifier_svm(train_set, train_label)
results = clf.evaluateResults(svm, test_set, test_label, estimator_name='SVM')
rbf = clf.rbf_classifier(train_set, train_label)
results = clf.evaluateResults(rbf, test_set, test_label, estimator_name='RBF')
# import pdb; pdb.set_trace()
random.shuffle(train_set)
Y_train = np.array([row[0] for row in train_set])
X_train = np.array([row[1] for row in train_set])
test_set = []
for key, value in test.items():
for file_id, words in value.items():
bow = np.zeros(len(vocab))
for word in words:
if word in vocab:
bow[vocab[word]] += 1
else:
bow[vocab['</unk>']] += 1
tf = bow/len(words)
tfidf = tf*idf_test
avg_glove = np.zeros(300)
for i in range(len(vocab)):
avg_glove += tfidf[i]*glove[i]
test_set.append([key, avg_glove])
print('shuffling test')
random.shuffle(test_set)
Y_test = np.array([row[0] for row in test_set])
X_test = np.array([row[1] for row in test_set])
print("Calling Classifiers\n")
print("Accuracy for Naive Bayes is : ", naive_bayes(X_train, Y_train, X_test, Y_test))
print("Accuracy for Logistic Regression is : ", logistic_regression(X_train, Y_train, X_test, Y_test))
print("Accuracy for SVM is : ", svm(X_train, Y_train, X_test, Y_test))
print("Accuracy for FF Neural Net is : ", fnn(X_train, Y_train, X_test, Y_test))
# print("Accuracy for Recurrent Neural Net is : ", rnn(X_train, Y_train, X_test, Y_t))
# Extract tweet partition
train_tweets = tweets[np.array(train)]
train_tweets, test_tweets, train_labels, test_labels = tweets[train], tweets[test], labels[train], labels[test]
print len(test_tweets)
print len(train_tweets)
train_tweets = np.hstack(train_tweets)
dictionary, tweets_features, vectorizer = bow.bow(train_tweets, vec="tfidf")
'''
Training different classifiers.
'''
svm = clf.classifier_svm(tweets_features, train_labels)
rf = clf.classifier_randomForest(tweets_features, train_labels)
ada = clf.adaboost(tweets_features, train_labels)
lr = clf.logistic_regression(tweets_features, train_labels)
'''
Test the different classifiers with the test tweets.
'''
pred = vectorizer.transform(test_tweets)
pred = pred.toarray()
_results, _accuracyLR, _precisionLR, _recallLR, _f_measureLR = clf.evaluateResults(lr, pred, test_labels,
estimator_name='Logistic regression',
file_name=results_folder)
_results, _accuracyRF, _precisionRF, _recallRF, _f_measureRF = clf.evaluateResults(rf, pred, test_labels,
estimator_name='RF',
file_name=results_folder)
reviews.append(review)
# length_of_reviews.append(len(review))
return reviews, scores #, length_of_reviews
if __name__ == '__main__':
print("OP_SPAM.py")
reviews, scores = parse_op_spam()
print(len(reviews), len(scores))
bow, vec = bow.bag_of_words(reviews)
# print(bow)
train_x, test_x, train_y, test_y = train_test_split(bow, scores, test_size=0.25, random_state=42)
classifier = classifiers.logistic_regression(train_x, train_y)
train_predictions = classifier.predict(train_x) # Training
train_accuracy = metrics.accuracy_score(train_y, train_predictions)
class_probabilities_train = classifier.predict_proba(train_x)
train_auc_score = metrics.roc_auc_score(train_y, class_probabilities_train[:, 1]);
print('\nTraining:')
print(' accuracy:', format(100 * train_accuracy, '.2f'))
print(' AUC value:', format(100 * train_auc_score, '.2f'))
test_predictions = classifier.predict(test_x) # Test
test_accuracy = metrics.accuracy_score(test_y, test_predictions)
class_probabilities_test = classifier.predict_proba(test_x)
test_auc_score = metrics.roc_auc_score(test_y, class_probabilities_test[:, 1]);
print('\nTesting:')
print(' accuracy:', format(100 * test_accuracy, '.2f'))
def main():
# Decision Tree
dtree_scores = decision_tree(5)
X, Y = extract_scores(dtree_scores)
tree_max = max(Y)
plt.plot(X, Y, 'r')
plt.axis([8, 22, 0.94, 0.98])
plt.title('Decision Tree Classifier Accuracy')
plt.xlabel('max_depth')
plt.ylabel('accuracy')
plt.savefig('dtree.png')
plt.clf()
# Logistic Regression
log_scores = logistic_regression(5)
X, Y = extract_scores(log_scores)
log_max = max(Y)
plt.plot(X, Y, 'r')
plt.axis([0, 1, 0.65, 0.8])
plt.title('Logistic Regression Accuracy')
plt.xlabel('alpha')
plt.ylabel('accuracy')
plt.savefig('log.png')
plt.clf()
# kNN
k_scores = kNN(5)
X, Y = extract_scores(k_scores)
k_max = max(Y)
plt.plot(X, Y, 'r')
plt.axis([0, 15, 0.95, 1.0])
plt.title('kNN Accuracy')
plt.xlabel('neighbors')
plt.ylabel('accuracy')
plt.savefig('knn.png')
plt.clf()
# neural network
net_scores = neural_network(5)
X, Y = extract_scores(net_scores)
net_max = max(Y)
plt.plot(X, Y, 'r')
plt.axis([0, 1.5, 0.95, 1.0])
plt.title('Neural Network Accuracy')
plt.xlabel('alpha')
plt.ylabel('accuracy')
plt.savefig('net.png')
plt.clf()
# Final Results
best = [tree_max, k_max, net_max]
models = ['DecisionTree', 'K-NN', 'NeuralNetwork']
x_pos = [i for i, _ in enumerate(models)]
plt.bar(x_pos, best, color='green')
plt.ylim(0.95, 1.0)
plt.xlabel("Classifiers")
plt.ylabel("Accuracy")
plt.title("Best Accuracy from Models")
plt.xticks(x_pos, models)
plt.savefig('best.png')
print('shuffling train')
random.shuffle(train_set)
Y_train = np.array([row[0] for row in train_set])
X_train = np.array([row[1] for row in train_set])
test_set = []
for key, value in test.items():
for file_id, words in value.items():
bow = np.zeros(len(vocab))
for word in words:
if word in vocab:
bow[vocab[word]] += 1
else:
bow[vocab['</unk>']] += 1
bow = bow / len(words)
test_set.append([key, bow])
print('shuffling test')
random.shuffle(test_set)
Y_test = np.array([row[0] for row in test_set])
X_test = np.array([row[1] for row in test_set])
print("Calling Classifiers\n")
print("Accuracy for Naive Bayes is : ",
naive_bayes(X_train, Y_train, X_test, Y_test))
print("Accuracy for Logistic Regression is : ",
logistic_regression(X_train, Y_train, X_test, Y_test))
print("Accuracy for SVM is : ", svm(X_train, Y_train, X_test, Y_test))
print("Accuracy for FF Neural Net is : ", fnn(X_train, Y_train, X_test,
Y_test))
# print("Accuracy for Recurrent Neural Net is : ", rnn(X_train, Y_train, X_test, Y_t))
train_tweets, test_tweets, train_labels, test_labels = tweets[
train], tweets[test], labels[train], labels[test]
print len(test_tweets)
print len(train_tweets)
train_tweets = np.hstack(train_tweets)
dictionary, tweets_features, vectorizer = bow.bow(train_tweets,
vec="tfidf")
'''
Training different classifiers.
'''
svm = clf.classifier_svm(tweets_features, train_labels)
rf = clf.classifier_randomForest(tweets_features, train_labels)
ada = clf.adaboost(tweets_features, train_labels)
lr = clf.logistic_regression(tweets_features, train_labels)
'''
Test the different classifiers with the test tweets.
'''
pred = vectorizer.transform(test_tweets)
pred = pred.toarray()
_results, _accuracyLR, _precisionLR, _recallLR, _f_measureLR = clf.evaluateResults(
lr,
pred,
test_labels,
estimator_name='Logistic regression',
file_name=results_folder)
_results, _accuracyRF, _precisionRF, _recallRF, _f_measureRF = clf.evaluateResults(
rf,
# print('Random Forest:')
# tree_count = clf.test_random_forest(analysis, 50, 10, True)
# print('Optimal Tree Count: {}.'.format(tree_count))
results = []
for i in np.round(np.linspace(0, 0.9, 10), 1).tolist():
print('Noise Amount: {}'.format(i))
# Preprocess
analysis = pr.Analyser(blur, oasis, i)
# analysis.get_summaries()
analysis.train_test()
analysis.get_tfidf()
# Fit logistic regression
print('Logistic Regression:')
logit_acc, logit_prec, logit_rec = clf.logistic_regression(
analysis, 10)
print('Accuracy: {}% +/-{}\nPrecsion: {}% +/- {}\nRecall: {}% +/- {}'.
format(np.round(logit_acc[0] * 100, 2),
np.round(100 * logit_acc[1], 2),
np.round(100 * logit_prec[0], 2),
np.round(100 * logit_prec[1], 2),
np.round(100 * logit_rec[0], 2),
np.round(100 * logit_rec[1], 2)))
result = ['logistic', i]
result.extend(logit_acc + logit_prec + logit_rec)
results.append(result)
print('{}\n'.format('-' * 80))
# Fit Random Forest
print('Random Forest:')
rf_acc, rf_prec, rf_rec = clf.random_forest(analysis, 100, 10)