parser.add_argument('--eval_batch_size', default=128, type=int)
parser.add_argument('--base_lr1', default=0.1, type=float)
parser.add_argument('--base_lr2', default=0.1, type=float)
parser.add_argument('--lr_factor', default=0.1, type=float)
parser.add_argument('--custom_weight_decay', default=5e-4, type=float)
parser.add_argument('--custom_momentum', default=0.9, type=float)
parser.add_argument('--load_ckpt_prefix', type=str, default='-')
parser.add_argument('--load_order', type=str, default='-')
parser.add_argument('--add_str', default=None, type=str)
the_args = parser.parse_args()
assert (the_args.nb_cl_fg % the_args.nb_cl == 0)
assert (the_args.nb_cl_fg >= the_args.nb_cl)
print(the_args)
np.random.seed(the_args.random_seed)
if not os.path.exists('./logs/cifar100_nfg50_ncls2_nproto20_mtl_exp01'):
print('Download checkpoints from Google Drive.')
os.system('sh ./script/download_ckpt.sh')
os.environ['CUDA_VISIBLE_DEVICES'] = the_args.gpu
print('Using gpu:', the_args.gpu)
occupy_memory(the_args.gpu)
print('Occupy GPU memory in advance.')
trainer = Trainer(the_args)
trainer.eval()
default='EBL',
choices=['HPL', 'EBL'])
parser.add_argument('--phase_sib',
type=str,
default='meta_train',
choices=['meta_train', 'meta_eval'])
parser.add_argument('--meta_eval_load_path',
type=str,
default='./ckpts/miniImageNet/e3bm_ckpt.pth')
args = parser.parse_args()
pprint(vars(args))
print('Experiment label: ' + args.label)
set_gpu(args.gpu)
occupy_memory(args.gpu)
print('Occupy GPU memory in advance.')
if args.baseline == 'MTL':
if args.seed == 0:
torch.backends.cudnn.benchmark = True
else:
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
if args.phase == 'meta_train':
trainer = MetaTrainer(args)
trainer.train()
elif args.phase == 'meta_eval':
print('Random mode.')
torch.backends.cudnn.benchmark = True
else:
import random
print('Fixed random seed:', args.seed)
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
torch.backends.cudnn.enabled = True
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
args.num_gpu = 1
if args.gpu_occupy:
occupy_memory(Config.gpu_id)
print('Occupy GPU memory in advance.')
trainer = MetaTrainer(args)
if args.mode == 'meta_train':
print('Start meta-train phase.')
trainer.train()
print('Start meta-test phase.')
trainer.eval()
elif args.mode == 'meta_eval':
print('Start meta-test phase.')
trainer.eval()
elif args.mode == 'pre_train':
print('Start pre-train phase.')
trainer.pre_train()