print("Running test for model: {}".format(args.m))
# Our input $x$
TEXT = torchtext.data.Field()
# Our labels $y$
LABEL = torchtext.data.Field(sequential=False, unk_token=None)
# Generate train/test splits from the SST dataset, filter out neutral examples
train, val, test = torchtext.datasets.SST.splits(
TEXT, LABEL,
filter_pred=lambda ex: ex.label != 'neutral')
TEXT.build_vocab(train)
LABEL.build_vocab(train)
train_iter, val_iter, test_iter = torchtext.data.BucketIterator.splits((train, val, test), batch_size=10, device=-1, repeat = False)
if args.m == "NaiveBayes":
alpha = 1
model = NaiveBayes(alpha, TEXT, LABEL)
model.train(train_iter, val_iter)
# Evaluate on training set
train_acc = model.evaluate(train_iter)
print('final train_acc (should be very high): ', train_acc)
# Evaluate on testing set
# test_code_NB(model, test_iter)
test_acc = model.evaluate(test_iter)
print('final test_acc: ', test_acc)
df['text'], df['is_spam'], test_size=0.2, random_state=191)
print('Data set:')
print('{} total'.format(df.shape[0]))
for t, t_name in zip(targets, target_names):
print('{} {}'.format(len(df[df['is_spam'] == t]), t_name))
print('\nTraining set:')
print('{} total'.format(len(X_train)))
for t, t_name in zip(targets, target_names):
print('{} {}'.format(sum([y == t for y in y_train]), t_name))
print('\nTest set:')
print('{} total'.format(len(X_test)))
for t, t_name in zip(targets, target_names):
print('{} {}'.format(sum([y == t for y in y_test]), t_name))
print('')
# Build Classifier
gvoc_model = NaiveBayes('General Vocabulary', X_train,
y_train, targets, target_names)
gvoc_model.train()
gvoc_model.evaluate(X_test, y_test, show_top_features=10)
rvoc_model = NaiveBayes('Reduced Vocabulary', X_train, y_train, targets,
target_names, max_features=200)
rvoc_model.train()
rvoc_model.evaluate(X_test, y_test, show_top_features=10)
def evaluate_naivebayes():
nb = NaiveBayes(sys.argv[1], evaluate = True)
out = nb.evaluate(sys.argv[2])
process(out, 'Naive Bayes')
