def test_all():
# Run on test data.
for test_subset in data['test']:
test_data = Data.BucketIterator(data['test'][test_subset],
args.batch_size,
args.cuda,
args,
shuffle=False)
print('=' * 89)
test_loss, test_acc, total_correctness = evaluate(test_data)
print('| subset %s | test loss %5.3f | test acc %8.3f' %
(test_subset, test_loss, test_acc))
print('=' * 89)
)
else:
torch.cuda.manual_seed(args.seed)
###############################################################################
# Load data
###############################################################################
dataset = torch.load(args.data)
vocab = dataset["vocab"]
data = dataset["data"]
ntokens = vocab.size()
if not args.test_only:
train_data = Data.BucketIterator(data['train'],
args.batch_size,
args.cuda,
args,
shuffle=True,
infor_weighting=args.infor_weighting)
valid_data = Data.BucketIterator(data['valid'],
args.batch_size,
args.cuda,
args,
shuffle=False)
test_data = Data.BucketIterator(data['test']["whole"],
args.batch_size,
args.cuda,
args,
shuffle=False)
if args.pre_trained is not None or args.embedding_file == "None":
embeddings = gen_embeddings(vocab, ntokens, args.emsize)
else: