def build_parser(self):
parser = get_parser("Augment final model of H^s-DAS config")
parser.add_argument('--name', required=True)
parser.add_argument('--dataset', type=str, default='imagenet', help='Imagenet')
parser.add_argument('--batch_size', type=int, default=64, help='batch size')
parser.add_argument('--lr', type=float, default=0.1, help='lr for weights')
parser.add_argument('--momentum', type=float, default=0.9, help='momentum')
parser.add_argument('--weight_decay', type=float, default=3e-5, help='weight decay')
parser.add_argument('--grad_clip', type=float, default=5.,
help='gradient clipping for weights')
parser.add_argument('--print_freq', type=int, default=200, help='print frequency')
parser.add_argument('--gpus', default='0', help='gpu device ids separated by comma. '
'`all` indicates use all gpus.')
parser.add_argument('--epochs', type=int, default=250, help='# of training epochs') # 600
parser.add_argument('--init_channels', type=int, default=36)
parser.add_argument('--layers', type=int, default=14, help='# of layers')
parser.add_argument('--seed', type=int, default=2, help='random seed')
parser.add_argument('--workers', type=int, default=4, help='# of workers')
parser.add_argument('--aux_weight', type=float, default=0.4, help='auxiliary loss weight')
parser.add_argument('--cutout_length', type=int, default=16, help='cutout length')
parser.add_argument('--drop_path_prob', type=float, default=0, help='drop path prob')
parser.add_argument('--gamma', type=float, default=0.97, help='learning rate decay')
parser.add_argument('--decay_period', type=int, default=1, help='epochs between two learning rate decays')
parser.add_argument('--genotype', required=True, help='Cell genotype')
parser.add_argument('--DAG', required=True, help='DAG genotype')
parser.add_argument('--dist', action='store_true', help='use multiprocess_distributed training')
parser.add_argument('--local_rank', default=0)
parser.add_argument('--resume_path', type=str, default=None)
parser.add_argument('--exclude_bias_and_bn', type=bool, default=True)
parser.add_argument('--warmup_epochs', type=int, default=10)
return parser
if self.args.likelihoodLearning:
self.likelihood = neural_likelihood(self)
if self.args.posteriorLearning:
self.posterior = neural_posterior(self)
# Define prior
self.prior = Prior(self.sim)
# Main loop
for round in range(self.args.numRound):
print(str(round)+"-th round of training start")
self.train(round)
def train(self, round):
# Batch of simulation input sampling
if self.args.likelihoodLearning:
self.thetas = sample_from_likelihood(self, self.args.simulation_budget_per_round, round == 0).detach().to(self.args.device)
else:
self.thetas = sample_from_posterior(self, self.args.simulation_budget_per_round, round == 0).detach().to(self.args.device)
# Simulation execution
self.simulated_output = self.sim.get_simulation_result(self.thetas)
print("simulation input, output shape : ", self.thetas.shape, self.simulated_output.shape)
# Posterior inference
train(self, round)
if __name__ == "__main__":
args = get_parser()
inference = main(args)
def build_parser(self):
# ======== cifar10 ============
parser = get_parser("Search cells of H-DAS config")
parser.add_argument('--name', required=True)
parser.add_argument('--dataset',
type=str,
default='cifar10',
help='CIFAR10')
parser.add_argument('--batch_size',
type=int,
default=64,
help='batch size')
parser.add_argument('--w_lr',
type=float,
default=0.025,
help='lr for weights')
parser.add_argument('--w_lr_min',
type=float,
default=0.001,
help='minimum lr for weights')
parser.add_argument('--w_momentum',
type=float,
default=0.9,
help='momentum for weights')
parser.add_argument('--w_weight_decay',
type=float,
default=3e-4,
help='weight decay for weights')
parser.add_argument('--w_grad_clip',
type=float,
default=5.,
help='gradient clipping for weights')
parser.add_argument('--print_freq',
type=int,
default=50,
help='print frequency')
parser.add_argument('--gpus',
default='0',
help='gpu device ids separated by comma. '
'`all` indicates use all gpus.')
parser.add_argument('--epochs',
type=int,
default=50,
help='# of training epochs')
parser.add_argument('--init_channels', type=int, default=16)
parser.add_argument('--layers',
type=int,
default=8,
help='# of layers')
parser.add_argument('--seed', type=int, default=2, help='random seed')
parser.add_argument('--workers',
type=int,
default=4,
help='# of workers')
parser.add_argument('--alpha_lr',
type=float,
default=3e-4,
help='lr for alpha')
parser.add_argument('--alpha_weight_decay',
type=float,
default=1e-3,
help='weight decay for alpha')
parser.add_argument('--local_rank', default=0)
parser.add_argument('--resume_path', type=str, default=None)
return parser
labels_np = np.concatenate(labels_list)
predict = np.argmax(score, axis=-1)
acc = np.sum(np.equal(labels_np, predict).astype(np.int32))/predict.shape[0]
if islog:
self._print_log('\tMean {} loss of {} batches: {}.'.format(
ln, len(self.data_loader[ln]), np.mean(loss_value)))
self._print_log('The accuracy over {} samples is {}'.format(predict.shape[0], acc))
else:
print('\tMean {} loss of {} batches: {}.'.format(
ln, len(self.data_loader[ln]), np.mean(loss_value)))
print('The accuracy over {} samples is {}'.format(predict.shape[0], acc))
if __name__ == '__main__':
parserfn = parser.get_parser()
# load arg form config file
print('Begin here')
p = parserfn.parse_args()
if p.config is not None:
# parse the yaml
with open(p.config, 'r') as f:
default_args = yaml.load(f)
key = vars(p).keys()
for k in default_args.keys():
if k not in key:
print('WRONG ARG: {}'.format(k))
assert (k in key)
