def init_config():
parser = myexman.ExParser(file=os.path.basename(__file__))
parser.add_argument('--name', type=str, default='')
parser.add_argument('--epochs', type=int, default=160)
parser.add_argument('--wd', type=float)
parser.add_argument('--lr_schedule', type=str, default='original')
parser.add_argument('--network', type=str, default='vgg')
parser.add_argument('--dataset', type=str, default='cifar10')
parser.add_argument('--batch_size', type=int, default=128)
parser.add_argument('--exception', type=int, nargs='*', default=[])
parser.add_argument('--ratio', type=float, default=0.9)
parser.add_argument('--prune_last', type=bool, default=True)
parser.add_argument('--run', type=str, default='')
parser.add_argument('--pruning', type=str, default='grasp')
parser.add_argument('--seed', type=int, default=0)
# Grasp params
parser.add_argument('--samples_per_class', type=int, default=10)
args = parser.parse_args()
return args
def main():
parser = myexman.ExParser(file=os.path.basename(__file__))
add_learner_params(parser)
is_help = False
if '--help' in sys.argv or '-h' in sys.argv:
sys.argv.pop(
sys.argv.index('--help' if '--help' in sys.argv else '-h'))
is_help = True
args, _ = parser.parse_known_args(log_params=False)
models.REGISTERED_MODELS[args.problem].add_model_hparams(parser)
if is_help:
sys.argv.append('--help')
args = parser.parse_args(namespace=args)
if args.data == 'imagenet' and args.aug == False:
raise Exception('ImageNet models should be eval with aug=True!')
if args.seed != -1:
random.seed(args.seed)
torch.manual_seed(args.seed)
cudnn.deterministic = True
args.gpu = 0
ngpus = torch.cuda.device_count()
args.number_of_processes = 1
if args.dist == 'ddp':
# add additional argument to be able to retrieve # of processes from logs
# and don't change initial arguments to reproduce the experiment
args.number_of_processes = args.world_size * ngpus
parser.update_params_file(args)
args.world_size *= ngpus
mp.spawn(
main_worker,
nprocs=ngpus,
args=(ngpus, args),
)
else:
parser.update_params_file(args)
main_worker(args.gpu, -1, args)
import utils
import utils_data
from torch import nn
from torch.optim.lr_scheduler import ReduceLROnPlateau
from models import ResNet_cifar100
from models import ResNet_cifar10
from models import ResNet_tinyimagenet
from models import ODENet_cifar100
from models import ODENet_cifar10
from models import ODENet_tinyimagenet
from logger import Logger
from utils import AverageMeter
parser = myexman.ExParser(file=os.path.basename(__file__))
parser.add_argument('--name', default='')
# Architecture
parser.add_argument('--time', type=eval, default=True)
parser.add_argument('--odenet', type=eval, default=False)
parser.add_argument('--n_blocks', type=int, default=2)
parser.add_argument('--norm', type=eval, default=False)
# Data
parser.add_argument('--data', default='cifar100', type=str)
parser.add_argument('--data_seed', default=30, type=int)
parser.add_argument('--val_size', default=0.2, type=float)
parser.add_argument('--train_bs', default=512, type=int)
parser.add_argument('--test_bs', default=512, type=int)
parser.add_argument('--augmentation', default=True, type=eval)
# Integration
parser.add_argument('--tol', type=float, default=1e-3)
(args.epochs - args.decrease_from) + 1, 1.)
return max(0, lr)
def predict(data, net):
pred = []
l = []
for x, y in data:
l.append(y.numpy())
x = x.to(device)
p = F.log_softmax(net(x), dim=1)
pred.append(p.data.cpu().numpy())
return np.concatenate(pred), np.concatenate(l)
parser = myexman.ExParser(file=__file__)
parser.add_argument('--data', default='cifar')
parser.add_argument('--num_examples', default=None, type=int)
parser.add_argument('--data_split_seed', default=42, type=int)
parser.add_argument('--resume', default='')
parser.add_argument('--lr',
default=0.001,
type=float,
help='Initial learning rate')
parser.add_argument('--gamma', default=0.5, type=float)
parser.add_argument('--epochs', default=100, type=int, help='Number of epochs')
parser.add_argument('--decrease_from', default=0, type=int)
parser.add_argument('--bs', default=256, type=int, help='Batch size')
parser.add_argument('--test_bs',
default=500,
type=int,