if __name__ == '__main__':
args = parse_args()
# Save vars
show_report = args.show_report
private_set = args.private_set
index = args.index
# Listing sorted checkpoints
ckpts = sorted(glob.glob(os.path.join(args.output, args.name, 'best*')),
reverse=True)
# Load original args
args = torch.load(ckpts[0])['args']
args = compute_args(args)
# Define the splits to be evaluated
evaluation_sets = ['valid', 'test'
] + ([private_set] if private_set is not None else [])
# Creating dataloader
train_dset = eval(args.dataloader)('train', args)
loaders = {
set: DataLoader(eval(args.dataloader)(set, args,
train_dset.token_to_ix),
args.batch_size,
num_workers=8,
pin_memory=True)
for set in evaluation_sets
}
type=str,
choices=['MELD', 'MOSEI', 'MOSI', 'IEMOCAP', 'VGAF'],
default='MOSEI')
parser.add_argument('--task',
type=str,
choices=['sentiment', 'emotion'],
default='sentiment')
parser.add_argument('--task_binary', type=bool, default=False)
args = parser.parse_args()
return args
if __name__ == '__main__':
# Base on args given, compute new args
args = compute_args(parse_args())
# Seed
torch.manual_seed(args.seed)
np.random.seed(args.seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
# DataLoader
train_dset = eval(args.dataloader)('train', args)
eval_dset = eval(args.dataloader)('valid', args, train_dset.token_to_ix)
train_loader = DataLoader(train_dset,
args.batch_size,
shuffle=True,
num_workers=8,