def test_train_and_test_dataset():
with tempfile.TemporaryDirectory() as temp_dir:
# Test both the normal cahce_dir and the given cache dir
for data_dir in [None, Path(temp_dir, 'twitter data')]:
train_data = wang_2017_election_twitter_train(data_dir)
test_data = wang_2017_election_twitter_test(data_dir)
assert len(train_data) > len(test_data)
combined_data = TargetTextCollection.combine(train_data, test_data)
assert 11899 == combined_data.number_targets()
model = AllenNLPModel(model_name, args.model_config, 'target-tagger',
args.model_save_dir)
if dataset_name == 'semeval_2014':
if not args.train_fp or not args.test_fp:
raise ValueError('If training and predicting for the SemEval '
'datasets the training and test file paths must '
'be given')
# As we are performing target extraction we use the conflict polarity
# targets like prior work
train_data = semeval_2014(args.train_fp, conflict=True)
test_data = semeval_2014(args.test_fp, conflict=True)
else:
temp_election_directory = Path('.', 'data', 'twitter_election_dataset')
train_data = wang_2017_election_twitter_train(temp_election_directory)
test_data = wang_2017_election_twitter_test(temp_election_directory)
if not args.model_save_dir.is_dir():
# Use the same size validation as the test data
test_size = len(test_data)
# Create the train and validation splits
train_data = list(train_data.values())
train_data, val_data = train_test_split(train_data,
test_size=test_size)
train_data = TargetTextCollection(train_data)
val_data = TargetTextCollection(val_data)
# Tokenize the data
datasets = [train_data, val_data, test_data]
tokenizer = spacy_tokenizer()
sizes = []
parser.add_argument("save_train_fp", type=parse_path,
help='File Path to save the new training dataset to')
parser.add_argument("save_val_fp", type=parse_path,
help='File Path to save the new validation dataset to')
parser.add_argument("save_test_fp", type=parse_path,
help='File Path to save the new test dataset to')
parser.add_argument("--conflict", action="store_true")
parser.add_argument("--random", action="store_true")
parser.add_argument("--remove_errors", action="store_true",
help=remove_errors_help)
args = parser.parse_args()
if args.dataset_name == 'election_twitter':
print(f'Downloading the Twitter dataset to {CACHE_DIRECTORY}')
train_dataset: TargetTextCollection = wang_2017_election_twitter_train()
test_dataset: TargetTextCollection = wang_2017_election_twitter_test()
else:
dataset_name_parser = {name: parser for name, parser in
zip(valid_dataset_names, dataset_parsers)}
dataset_parser = dataset_name_parser[args.dataset_name]
train_dataset: TargetTextCollection = dataset_parser(args.train_dataset_fp,
conflict=args.conflict)
test_dataset: TargetTextCollection = dataset_parser(args.test_dataset_fp,
conflict=args.conflict)
if args.dataset_name == 'semeval_2016':
train_dataset = train_dataset.one_sample_per_span(remove_empty=True)
test_dataset = test_dataset.one_sample_per_span(remove_empty=True)
# If the task is sentiment prediction remove all of the sentences that
# do not have targets
if args.task == 'sentiment':
import tempfile
from pathlib import Path
from time import time
from target_extraction.dataset_parsers import wang_2017_election_twitter_train, wang_2017_election_twitter_test
with tempfile.TemporaryDirectory() as temp_dir:
another = Path(temp_dir, 'first')
t = time()
a = wang_2017_election_twitter_train(another)
a_t = a.number_targets()
b = wang_2017_election_twitter_test(another)
b_t = b.number_targets()
print('done')