def __init__(self, model):
model.eval()
self.model = model
self.correct = 0
class Stat:
def __init__(self, d, dnm):
self.domain = d
self.name = dnm
self.width = 0
self.max_eps = 0
self.safe = 0
self.proved = 0
self.time = 0
self.domains = [ Stat(h.parseValues(domains,d), h.catStrs(d)) for d in args.test_domain ]
default=None,
help='use regularization')
parser.add_argument("--gpu_id",
type=str,
default=None,
help="specify gpu id, None for all")
parser.add_argument("--decay-fir",
type=bool,
default=False,
help="decay the first Mix domain")
args = parser.parse_args()
if args.gpu_id is not None:
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu_id
largest_domain = max([len(h.catStrs(d)) for d in (args.domain)])
largest_test_domain = max([len(h.catStrs(d)) for d in (args.test_domain)])
args.log_interval = int(50000 / (args.batch_size * args.log_freq))
h.max_c_for_norm = args.max_norm
if h.use_cuda:
torch.cuda.manual_seed(1 + args.seed)
else:
torch.manual_seed(args.seed)
train_loader = h.loadDataset(args.dataset, args.batch_size, True, False)
val_loader = h.loadDataset(args.dataset, args.batch_size, True, False, True)
test_loader = h.loadDataset(args.dataset, args.test_batch_size, False, False)
parser.add_argument('-n', '--net', choices = h.getMethodNames(models), action = 'append'
, default=[], help='picks which net to use for training') # one net for now
parser.add_argument('-D', '--dataset', choices = [n for (n,k) in inspect.getmembers(datasets, inspect.isclass) if issubclass(k, Dataset)]
, default="MNIST", help='picks which dataset to use.')
parser.add_argument('-o', '--out', default="out/", help='picks which net to use for training')
parser.add_argument('--dont-write', type=h.str2bool, nargs='?', const=True, default=False, help='dont write anywhere if this flag is on')
parser.add_argument('--test-size', type=int, default=2000, help='number of examples to test with')
parser.add_argument('-r', '--regularize', type=float, default=None, help='use regularization')
args = parser.parse_args()
largest_domain = max([len(h.catStrs(d)) for d in (args.domain)] )
largest_test_domain = max([len(h.catStrs(d)) for d in (args.test_domain)] )
args.log_interval = int(50000 / (args.batch_size * args.log_freq))
h.max_c_for_norm = args.max_norm
if h.use_cuda:
torch.cuda.manual_seed(1 + args.seed)
else:
torch.manual_seed(args.seed)
train_loader = h.loadDataset(args.dataset, args.batch_size, True, False)
test_loader = h.loadDataset(args.dataset, args.test_batch_size, False, False)
input_dims = train_loader.dataset[0][0].size()