def infer(root_path, top_k=1):
if C.get()['mode'] != 'test':
return _infer(model, root_path)
if C.get()['infer_mode'] == 'ret':
return _infer_ret(model, root_path)
else:
return _infer(model, root_path)
def test_git_info():
config = Config()
if '_git' in config:
assert 'wbaek/theconf.git' in config['_git']['remote']
Config.clear()
def new_forward(self, x, extract=False):
if isinstance(self, DataParallel):
self = self.module
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
feat2 = torch.flatten(x, 1)
x = self.avgpool(x)
x = torch.flatten(x, 1)
feat = x
x = self.fc(x)
x = softmax(x, dim=1)
if C.get()['feat'] == 'feat2':
return feat2
elif C.get()['feat'] == 'pool':
return feat
else:
return x
def test_contains():
config = Config()
config['key'] = 1
assert ('key' in config) == True
assert ('not_exists' in config) == False
Config.clear()
def _infer(model, root_path, test_loader=None):
if test_loader is None:
test_loader = data_loader(root=os.path.join(root_path, 'test_data'), phase='test')
model.eval()
res_fc = None
res_id = None
for idx, (data_id, image, _) in enumerate(tqdm(test_loader)):
image = image.cuda()
with torch.no_grad():
fc = model(image)
fc = fc.detach().cpu().numpy()
fc = np_softmax(fc)
# with torch.no_grad():
# fc2 = model(torch.flip(image, (3, ))) # TTA : horizontal flip
# fc2 = fc2.detach().cpu().numpy()
# fc2 = np_softmax(fc2)
# fc = fc + fc2
if C.get()['infer_mode'] == 'face':
fc[:, range(60)] = -1
# target_lb = list(range(60, 100))
if idx == 0:
res_fc = fc
res_id = data_id
else:
res_fc = np.concatenate((res_fc, fc), axis=0)
res_id = res_id + data_id
res_cls = np.argmax(res_fc, axis=1)
return [res_id, res_cls]
def test_get_default():
config = Config()
config['key'] = 1
assert config.get('key', 0) == 1
assert config.get('not_exists', 0) == 0
Config.clear()
def get_transform(random_crop=True):
normalize = transforms.Normalize(
mean=[x / 255.0 for x in [125.3, 123.0, 113.9]],
std=[x / 255.0 for x in [63.0, 62.1, 66.7]]
)
transform = []
if random_crop:
aug = C.get()['aug']
if aug == 'faa_imagenet':
transform.append(Augmentation(fa_reduced_imagenet()))
elif aug == 'faa_cifar10':
transform.append(Augmentation(fa_reduced_cifar10()))
elif aug == 'rot360':
transform.append(transforms.RandomRotation(360))
elif aug == 'default':
pass
resize_method = C.get().conf.get('resize', 'ResizedCrop')
if resize_method == 'ResizedCrop':
transform.append(transforms.RandomResizedCrop(224, interpolation=Image.LANCZOS))
elif resize_method == 'Resize':
transform.append(transforms.Resize(256, interpolation=Image.LANCZOS))
transform.append(transforms.CenterCrop(224))
else:
raise ValueError(resize_method)
# transform.append(transforms.RandomResizedCrop(224, scale=(0.8, 1.0), ratio=(0.8, 1.2), interpolation=Image.LANCZOS))
transform.append(transforms.RandomHorizontalFlip())
transform.append(transforms.RandomRotation(20))
else:
transform.append(transforms.Resize(256, interpolation=Image.LANCZOS))
transform.append(transforms.CenterCrop(224))
transform.append(transforms.ToTensor())
transform.append(normalize)
return transforms.Compose(transform)
def test_flatten(datafiles):
filenames = [str(f) for f in datafiles.listdir()]
config = Config(filenames[0], skip_timestamp=True, skip_git_info=True)
config['tar'] = 'test'
config['var'] = 'variation'
p = config._flatten([], config.conf)
assert len(p) == 4
assert p[0] == ('foo-bar', 1)
assert p[1] == ('foo-baz', 2)
assert p[2] == ('tar', 'test')
assert p[3] == ('var', 'variation')
p = config.flatten()
assert len(p) == 4
assert p['foo-bar'] == 1
assert p['foo-baz'] == 2
assert p['tar'] == 'test'
assert p['var'] == 'variation'
Config.clear()
def test_update_and_dump(datafiles):
filenames = [str(f) for f in datafiles.listdir()]
config = Config(filenames[0], skip_timestamp=True, skip_git_info=True)
assert config['_version'] == 1
assert config['foo']['bar'] == 1
assert config['foo']['baz'] == 2
config['foo']['bar'] = {'test': 10}
config['foo']['baz'] = 3
assert config['foo']['bar']['test'] == 10
assert config['foo']['baz'] == 3
yaml_string = config.dump()
assert yaml_string == '_version: 2\nfoo:\n bar:\n test: 10\n baz: 3\n'
yaml_string = config.dump()
assert yaml_string == '_version: 3\nfoo:\n bar:\n test: 10\n baz: 3\n'
config.dump(filenames[0])
Config.clear()
config = Config(filenames[0])
assert config['foo']['bar']['test'] == 10
assert config['foo']['baz'] == 3
Config.clear()
def test_singletone2(datafiles):
filenames = [str(f) for f in datafiles.listdir()]
_ = Config(filenames[0])
with pytest.raises(Exception) as _:
_ = Config(filenames[0])
Config.clear()
def test_singletone():
conf1 = Config.get_instance()
conf2 = Config.get_instance()
conf3 = Config.get()
assert conf1 == conf2
Config.clear()
def test_timestamp():
config = Config()
assert datetime.datetime.now().strftime('%Y/%m/%d') in config['_timestamp']
Config.clear()
# mode argument
args = ConfigArgumentParser(conflict_handler='resolve')
args.add_argument("--cv", type=int, default=0)
args.add_argument("--ratio", type=float, default=0.1)
# reserved for nsml
args.add_argument("--cuda", type=bool, default=True)
args.add_argument("--mode", type=str, default="train")
args.add_argument("--iteration", type=str, default='0')
args.add_argument("--pause", type=int, default=0)
args.add_argument("--transfer", type=bool, default=False)
config = args.parse_args()
logger.info(str(C.get().conf))
num_classes = C.get()['num_class']
base_lr = config.lr
cuda = config.cuda
eval_split = 'val'
mode = config.mode
if C.get()['model'] == 'resnet18':
model = models.resnet18(pretrained=None)
model.avgpool = nn.AdaptiveAvgPool2d((1, 1))
model.fc = nn.Linear(512 * 1, num_classes)
elif C.get()['model'] == 'resnet50':
model = models.resnet50(pretrained=None)
model.avgpool = nn.AdaptiveAvgPool2d((1, 1))
model.fc = nn.Linear(512 * 4, num_classes)
if __name__ == '__main__':
# mode argument
args = ConfigArgumentParser(conflict_handler='resolve')
args.add_argument("--cv", type=int, default=0)
args.add_argument("--ratio", type=float, default=0.1)
# reserved for nsml
args.add_argument("--cuda", type=bool, default=True)
args.add_argument("--mode", type=str, default="train")
args.add_argument("--iteration", type=str, default='0')
args.add_argument("--pause", type=int, default=0)
config = args.parse_args()
logger.info(str(C.get().conf))
num_classes = C.get()['num_class']
base_lr = config.lr
cuda = config.cuda
eval_split = 'val'
mode = config.mode
model = models.resnet50(pretrained=None)
model.avgpool = nn.AdaptiveAvgPool2d((1, 1))
model.fc = nn.Linear(512 * 4, num_classes)
loss_fn = nn.CrossEntropyLoss()
if cuda:
model = model.cuda()
loss_fn = loss_fn.cuda()
