use_gpu = args.gpu > 0 and torch.cuda.device_count() > 0
if use_gpu and args.device:
print('Select GPU {}'.format(args.device))
torch.cuda.set_device(args.device)
assert mode in ['train', 'test', 'predict'], 'Unknown mode: {}'.format(mode)
# ----------------------------------------------------------------------
# Load data
if mode == 'train':
train_set = Dataset(train_file, labels=labels)
dev_set = Dataset(dev_file, labels=labels)
test_set = Dataset(test_file, labels=labels)
train_token_count, train_label_count = train_set.data_stats()
dev_token_count, dev_label_count = dev_set.data_stats()
test_token_count, test_label_count = test_set.data_stats()
token_vocab = {'$UNK$': 0}
label_vocab = {'NM': 0}
for t in list(train_token_count.keys()) \
+ list(dev_token_count.keys()) + list(test_token_count.keys()):
if t not in token_vocab:
token_vocab[t] = len(token_vocab)
for l in list(train_label_count.keys()) \
+ list(dev_label_count.keys()) + list(test_label_count.keys()):
if l not in label_vocab:
label_vocab[l] = len(label_vocab)
train_set.token_vocab = token_vocab
