def getTestDataset(test_dataset='cifar10', batch_size=32):
if test_dataset == 'Imagenet':
testFolder = os.path.join(os.path.dirname(os.path.abspath(__file__)),
'../val/')
print('testFolder', testFolder)
Imagenet = datasets.ImageNet12(trainFolder=None, testFolder=testFolder)
test_loader = Imagenet.getTestLoader(batch_size=batch_size,
num_workers=0)
else:
datasets.BASE_DATA_FOLDER = os.path.join(
os.path.dirname(os.path.abspath(__file__)), '../')
try:
os.mkdir(datasets.BASE_DATA_FOLDER)
except:
pass
cifar10 = datasets.CIFAR10()
test_loader = cifar10.getTestLoader(batch_size=batch_size,
num_workers=0)
return test_loader
try:
os.mkdir(imageNet12modelsFolder)
except:
pass
print('Batch size: {}'.format(batch_size))
if batch_size % NUM_GPUS != 0:
raise ValueError(
'Batch size: {} must be a multiple of the number of gpus:{}'.format(
batch_size, NUM_GPUS))
imageNet12 = datasets.ImageNet12('...',
'...',
type_of_data_augmentation='extended',
already_scaled=False,
pin_memory=True)
train_loader = imageNet12.getTrainLoader(batch_size, shuffle=True)
test_loader = imageNet12.getTestLoader(batch_size, shuffle=False)
# # Teacher model
alexnet_unquantized = torchvision.models.alexnet(pretrained=True)
if USE_CUDA:
alexnet_unquantized = alexnet_unquantized.cuda()
if NUM_GPUS > 1:
alexnet_unquantized = torch.nn.parallel.DataParallel(alexnet_unquantized)
#Train a wide-resNet with quantized distillation
quant_distilled_model_name = 'alexnet_quant_distilled{}bits'.format(NUM_BITS)
def __init__(self, args):
self.criterion = nn.CrossEntropyLoss().cuda()
self.init_channels = args.init_channels
self.num_blocks = args.num_blocks
self.momentum = args.momentum
self.weight_decay = args.weight_decay
self.learning_rate = args.learning_rate
self.grad_clip = args.grad_clip
self.epochs = args.epochs_step
#self.device = torch.device("cuda:"+args.gpu[0])
self.save = args.save
self.report_freq = args.report_freq
self.args = args
self.target_param = 0.2 # mystery 改成初始化模型的平均值
self.bucket_size = 256
num_workers = 1
pin_memory = False
shuffle = True
self.share_params = {}
self.search_space = args.search_space
if args.dataset == 'cifar10':
root = os.path.join(os.path.dirname(os.path.abspath(__file__)),
args.data)
print(root)
train_transform, valid_transform = utils._data_transforms_cifar10(
args)
train_data = dset.CIFAR10(root=root,
train=True,
download=False,
transform=train_transform)
test_data = dset.CIFAR10(root=root,
train=False,
download=False,
transform=valid_transform)
self.trainloader = torch.utils.data.DataLoader(
train_data,
batch_size=args.batch_size,
shuffle=shuffle,
num_workers=num_workers,
pin_memory=pin_memory)
self.testloader = torch.utils.data.DataLoader(
test_data,
batch_size=args.batch_size,
shuffle=shuffle,
num_workers=num_workers,
pin_memory=pin_memory)
elif args.dataset == 'Imagenet':
trainFolder = os.path.join(args.data, 'train/')
testFolder = os.path.join(args.data, 'val/')
print('testFolder', testFolder)
Imagenet = datasets.ImageNet12(trainFolder=trainFolder,
testFolder=testFolder)
self.trainloader = Imagenet.getTrainLoader(
batch_size=args.batch_size, num_workers=56)
print('train data done!')
self.testloader = Imagenet.getTestLoader(
batch_size=args.batch_size, num_workers=56)
print('test data done!')
else:
raise ValueError('Unknown dataset {}'.format(args.dataset))