def run(task_name, data_dir, pipeline_name, print_predictions, error_analysis,
remove_stopwords):
logger.info('>> Running {} experiment'.format(task_name))
tsk = task(task_name)
logger.info('>> Loading data...')
tsk.load(data_dir)
logger.info('>> retrieving train/data instances...')
train_X, train_y, test_X, test_y = utils.get_instances(
tsk, split_train_dev=False)
logger.info('>> Descriptive statistics dataset:')
utils.descriptive_statistics(train_X, train_y, test_X, test_y)
test_X_ref = test_X
if remove_stopwords:
if pipeline_name.startswith('cnn'):
pipeline_name = pipeline_name.split('_')[0]
pipeline_name = pipeline_name + '_stopwords'
if pipeline_name.startswith('cnn'):
pipe = cnn(pipeline_name)
train_X, train_y, test_X, test_y = pipe.encode(train_X, train_y,
test_X, test_y)
logger.info('>> testing CNN...')
else:
pipe = pipeline(pipeline_name)
logger.info('>> training pipeline ' + pipeline_name)
pipe.fit(train_X, train_y)
if pipeline_name == 'naive_bayes_counts_lex':
logger.info(" -- Found {} tokens in lexicon".format(
pipe.tokens_from_lexicon))
logger.info('>> testing...')
sys_y = pipe.predict(test_X)
# logger.info(utils.print_prediction(test_X, test_y, sys_y))
if print_predictions:
logger.info('>> predictions1')
utils.print_all_predictions(test_X_ref, test_y, sys_y, logger)
if error_analysis:
# Used for error evaluation
logger.info(utils.print_error_analysis(test_X, test_y, sys_y))
# logger.info(utils.print_confusion_matrix(test_y, sys_y)) # Prints the confusion matrix
utils.eval(test_y, sys_y, pipeline_name, data_dir)
if pipeline_name.startswith('naive_bayes'):
utils.important_features_per_class(pipe.named_steps.frm,
pipe.named_steps.clf,
n=10)
def run(task_name, data_dir, pipeline_name, print_predictions):
logger.info('>> Running {} experiment'.format(task_name))
tsk = task(task_name)
logger.info('>> Loading data...')
tsk.load(data_dir)
logger.info('>> retrieving train/data instances...')
train_X, train_y, test_X, test_y = utils.get_instances(
tsk, split_train_dev=False)
test_X_ref = test_X
if pipeline_name.startswith('cnn'):
pipe = cnn(pipeline_name)
train_X, train_y, test_X, test_y = pipe.encode(train_X, train_y,
test_X, test_y)
logger.info('>> testing...')
else:
pipe = pipeline(pipeline_name)
logger.info('>> training pipeline ' + pipeline_name)
pipe.fit(train_X, train_y)
if pipeline_name == 'naive_bayes_counts_lex':
logger.info(" -- Found {} tokens in lexicon".format(
pipe.tokens_from_lexicon))
logger.info('>> testing...')
sys_y = pipe.predict(test_X)
logger.info('>> evaluation...')
logger.info(utils.eval(test_y, sys_y))
if print_predictions:
logger.info('>> predictions')
utils.print_all_predictions(test_X_ref, test_y, sys_y, logger)
def run(task_name, data_dir, pipeline_name):
logger.info('>> Running {} experiment'.format(task_name))
tsk = task(task_name)
logger.info('>> Loading data...')
tsk.load(data_dir)
logger.info('>> retrieving train/test instances...')
train_X, train_y, test_X, test_y = utils.get_instances(
tsk, split_train_dev=False)
if pipeline_name.startswith('cnn'):
pipe = cnn(pipeline_name)
train_X, train_y, test_X, test_y = pipe.encode(train_X, train_y,
test_X, test_y)
logger.info('>> testing...')
else:
pipe = pipeline(pipeline_name)
logger.info('>> training pipeline ' + pipeline_name)
pipe.fit(train_X, train_y)
logger.info('>> testing...')
sys_y = pipe.predict(test_X)
logger.info(utils.print_prediction(test_X, test_y, sys_y))
logger.info('>> evaluation...')
logger.info(utils.eval(test_y, sys_y))
def evaluate(model, test_X, test_y):
#sys_y = model.predict_classes(test_X, batch_size=128)
sys_y = (model.predict(test_X) > 0.5).astype('int32')
#sys_y = model.predict(test_X > 0.5).astype('int32')
print(utils.eval(test_y, sys_y))
def evaluate(model, test_X, test_y):
sys_y = model.predict_classes(test_X, batch_size=128)
print(utils.eval(test_y, sys_y))