def __init__(self, network, learning_rate):
"""
Load Data, initialize a given network structure and set learning rate
DO NOT MODIFY
"""
# Define a transform to normalize the data
transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize((0.5,), (0.5,))])
# Download and load the training data
trainset = datasets.KMNIST(root='./data', train=True, download=True, transform=transform)
self.trainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=False)
# Download and load the test data
testset = datasets.KMNIST(root='./data', train=False, download=True, transform=transform)
self.testloader = torch.utils.data.DataLoader(testset, batch_size=64, shuffle=False)
self.model = network
"""
TODO: Set appropriate loss function such that learning is equivalent to minimizing the
cross entropy loss. Note that we are outputting log-softmax values from our networks,
not raw softmax values, so just using torch.nn.CrossEntropyLoss is incorrect.
Hint: All networks output log-softmax values (i.e. log probabilities or.. likelihoods.).
"""
self.lossfn = nn.NLLLoss()
self.optimizer = optim.Adam(self.model.parameters(), lr=learning_rate)
self.num_train_samples = len(self.trainloader)
self.num_test_samples = len(self.testloader)
def setup_data_loaders(self):
if self.dataset == 'celeba':
transform_list = [transforms.CenterCrop(140), transforms.Resize((64,64),PIL.Image.ANTIALIAS), transforms.ToTensor()]
#if self.input_normalize_sym:
#D = 64*64*3
#transform_list.append(transforms.LinearTransformation(2*torch.eye(D), -.5*torch.ones(D)))
transform = transforms.Compose(transform_list)
train_dataset = datasets.CelebA(self.datadir, split='train', target_type='attr', download=True, transform=transform)
test_dataset = datasets.CelebA(self.datadir, split='test', target_type='attr', download=True, transform=transform)
self.nlabels = 0
elif self.dataset == 'mnist':
train_dataset = datasets.MNIST(self.datadir, train=True, target_transform=None, download=True, transform=transforms.Compose([transforms.ToTensor()]))
test_dataset = datasets.MNIST(self.datadir, train=False, target_transform=None, download=True, transform=transforms.Compose([transforms.ToTensor()]))
self.nlabels = 10
elif self.dataset == 'fashionmnist':
train_dataset = datasets.FashionMNIST(self.datadir, train=True, target_transform=None, download=True, transform=transforms.Compose([transforms.ToTensor()]))
test_dataset = datasets.FashionMNIST(self.datadir, train=False, target_transform=None, download=True, transform=transforms.Compose([transforms.ToTensor()]))
self.nlabels = 10
elif self.dataset == 'kmnist':
train_dataset = datasets.KMNIST(self.datadir, train=True, target_transform=None, download=True, transform=transforms.Compose([transforms.ToTensor()]))
test_dataset = datasets.KMNIST(self.datadir, train=False, target_transform=None, download=True, transform=transforms.Compose([transforms.ToTensor()]))
self.nlabels = 10
else:
raise Exception("Dataset not found: " + dataset)
if self.limit_train_size is not None:
train_dataset = torch.utils.data.random_split(train_dataset, [self.limit_train_size, len(train_dataset)-self.limit_train_size])[0]
self.train_loader = torch.utils.data.DataLoader(DatasetWithIndices(train_dataset, self.input_normalize_sym), batch_size=self.batch_size, shuffle=True, **self.dataloader_kwargs)
self.test_loader = torch.utils.data.DataLoader(DatasetWithIndices(test_dataset, self.input_normalize_sym), batch_size=self.batch_size, shuffle=False, **self.dataloader_kwargs)
def load_kmnist(path="~/.datasets/",
train_test_split=0.8,
transform=transforms.ToTensor(),
random_state=42):
"""Loads the Kuzushiji-MNIST dataset.
Args:
path (str): Path for the files.
Returns:
train_dataset, test_dataset, validation_dataset
"""
torch.manual_seed(random_state)
ds = datasets.KMNIST(root=path,
train=True,
download=True,
transform=transform)
test_ds = datasets.KMNIST(root=path,
train=False,
download=True,
transform=transform)
train_size = int(train_test_split * len(ds))
test_size = len(ds) - train_size
train_ds, val_ds = torch.utils.data.random_split(ds,
[train_size, test_size])
return train_ds, val_ds, test_ds
def get_data():
data_dir = Path("/media/wwymak/Storage/kmnist")
data_dir.mkdir(exist_ok=True)
kmnist = datasets.KMNIST(data_dir, download=True)
kmnist_test = datasets.KMNIST(data_dir, train=False)
X_train = kmnist.data
y_train = kmnist.targets
X_test = kmnist_test.data
y_test = kmnist_test.targets
X_train = X_train.view(-1, 28 * 28).float()
X_test = X_test.view(-1, 28 * 28).float()
# train_mean = X_train.mean()
# train_std = X_train.std()
#
# X_train = normalise(X_train, train_mean, train_std)
# X_test = normalise(X_test, train_mean, train_std)
#
# X_train, X_test, y_train, y_test = map(from_numpy, (X_train, X_test, y_train, y_test))
# X_train, X_test = [x.float() for x in (X_train, X_test)]
# y_train, y_test = [x.long() for x in (y_train, y_test)]
#
# X_train = X_train.reshape([X_train.shape[0], -1])
# X_test = X_test.reshape([X_test.shape[0], -1])
return X_train, X_test, y_train, y_test
def __init__(self, network, learning_rate):
"""
Load Data, initialize a given network structure and set learning rate
"""
# Define a transform to normalize the data
transform = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize((0.5, ), (0.5, ))])
# Download and load the training data
trainset = datasets.KMNIST(root='./data',
train=True,
download=True,
transform=transform)
self.trainloader = torch.utils.data.DataLoader(trainset,
batch_size=64,
shuffle=False)
# Download and load the test data
testset = datasets.KMNIST(root='./data',
train=False,
download=True,
transform=transform)
self.testloader = torch.utils.data.DataLoader(testset,
batch_size=64,
shuffle=False)
self.model = network
self.lossfn = torch.nn.NLLLoss()
self.optimizer = optim.Adam(self.model.parameters(), lr=learning_rate)
self.num_train_samples = len(self.trainloader)
self.num_test_samples = len(self.testloader)
def get_KMNIST(
data_dir,
image_size=28,
mean=(0.19176215,), std=(0.33852664,),
train_transform=None, test_transform=None,
download=False
):
if train_transform is None:
train_transform = transforms.Compose([
transforms.RandomCrop(image_size, padding=4),
transforms.RandomAffine(degrees=45, translate=(0.1, 0.1), scale=(0.8, 1.2)),
transforms.ToTensor(),
transforms.Normalize(mean, std)
])
if test_transform is None:
test_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean, std)
])
in_channels = 1
classes = 10
dataset = datasets.KMNIST(data_dir, train=True, download=download, transform=train_transform)
train_indices, validation_indices = split_train_and_validation_datasets(dataset, classes)
train_dataset = KMNIST(data_dir, train_indices, train=True, download=download, transform=train_transform)
validation_dataset = KMNIST(data_dir, validation_indices, train=True, download=download, transform=test_transform)
test_dataset = datasets.KMNIST(data_dir, train=False, download=download, transform=test_transform)
return train_dataset, validation_dataset, test_dataset, in_channels, classes
def get_dataset(dataset_name, model_name):
train_dataset = []
test_dataset = []
trans = transforms.ToTensor()
if model_name == "cnn" or model_name == "cnv":
trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])
if dataset_name == 'mnist':
train_dataset = dsets.MNIST(root='./data',
train=True,
transform=trans,
download=True)
test_dataset = dsets.MNIST(root='./data',
train=False,
transform=trans)
elif dataset_name == "kmnist":
train_dataset = dsets.KMNIST(root='./data',
train=True,
transform=trans,
download=True)
test_dataset = dsets.KMNIST(root='./data',
train=False,
transform=trans)
return train_dataset, test_dataset
def __init__(self, network, learning_rate):
"""
Load Data, initialize a given network structure and set learning rate
DO NOT MODIFY
"""
# Define a transform to normalize the data
transform = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize((0.5, ), (0.5, ))])
# Download and load the training data
trainset = datasets.KMNIST(root='./data',
train=True,
download=True,
transform=transform)
self.trainloader = torch.utils.data.DataLoader(trainset,
batch_size=64,
shuffle=False)
# Download and load the test data
testset = datasets.KMNIST(root='./data',
train=False,
download=True,
transform=transform)
self.testloader = torch.utils.data.DataLoader(testset,
batch_size=64,
shuffle=False)
self.model = network
"""
TODO: Set appropriate loss function such that learning is equivalent to minimizing the
cross entropy loss. Note that we are outputting log-softmax values from our networks,
not raw softmax values, so just using torch.nn.CrossEntropyLoss is incorrect.
Hint: All networks output log-softmax values (i.e. log probabilities or.. likelihoods.).
"""
# self.lossfn = lambda output, label: {
# # 64,10 -> 64
# # all batch?
# }
# self.lossfn = lambda outputs, target: outputs[range(target.shape[0]), target].sum()
# self.lossfn = lambda outputs, target: -1*outputs[range(target.shape[0]), target].mean()
self.lossfn = F.nll_loss # this seems good
# self.lossfn = torch.nn.NLLLoss() # difference?
# or:
# note: the cross entropy loss already includes the softmax
# so the following loss functions are not correct
# self.nNn = torch.nn.CrossEntropyLoss()
# self.lossfn = lambda outputs, target: self.nNn(torch.exp(outputs), target)
# or: try this
# self.lossfn = lambda outputs, target: (torch.nn.CrossEntropyLoss())(torch.exp(outputs), target)
# mean or sum? positive or negative?
self.optimizer = optim.Adam(self.model.parameters(), lr=learning_rate)
self.num_train_samples = len(self.trainloader)
self.num_test_samples = len(self.testloader)
def get_kmnist_loaders(arguments):
transformers = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))])
train_set = datasets.KMNIST(DATASET_PATH,
train=True,
download=True,
transform=transformers)
test_set = datasets.KMNIST(DATASET_PATH,
train=False,
download=True,
transform=transformers)
return load(arguments, test_set, train_set)
def main():
# command-line arguments
parser = argparse.ArgumentParser()
parser.add_argument('--net', type=str, default='full',
help='lin, full or conv')
parser.add_argument('--lr', type=float, default=0.01, help='learning rate')
parser.add_argument('--mom', type=float, default=0.5, help='momentum')
parser.add_argument('--epochs', type=int, default=10,
help='number of training epochs')
parser.add_argument('--no_cuda', action='store_true',
default=False, help='disables CUDA')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
device = torch.device('cuda' if use_cuda else 'cpu')
kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
# define a transform to normalize the data
transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize((0.5,), (0.5,))])
# fetch and load training data
trainset = datasets.KMNIST(
root='./data', train=True, download=True, transform=transform)
train_loader = torch.utils.data.DataLoader(
trainset, batch_size=64, shuffle=False)
# fetch and load test data
testset = datasets.KMNIST(
root='./data', train=False, download=True, transform=transform)
test_loader = torch.utils.data.DataLoader(
testset, batch_size=64, shuffle=False)
# choose network architecture
if args.net == 'lin':
net = NetLin().to(device)
elif args.net == 'full':
net = NetFull().to(device)
else:
net = NetConv().to(device)
if list(net.parameters()):
# use SGD optimizer
optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=args.mom)
# training and testing loop
for epoch in range(1, args.epochs + 1):
train(args, net, device, train_loader, optimizer, epoch)
test(args, net, device, test_loader)
def get_dataset(self):
"""
Uses torchvision.datasets.KMNIST to load dataset.
Downloads dataset if doesn't exist already.
Returns:
torch.utils.data.TensorDataset: trainset, valset
"""
trainset = datasets.KMNIST('datasets/KMNIST/train/', train=True, transform=self.train_transforms,
target_transform=None, download=True)
valset = datasets.KMNIST('datasets/KMNIST/test/', train=False, transform=self.val_transforms,
target_transform=None, download=True)
return trainset, valset
def load_data(args, shuffle, droplast):
# Transform to grayscale
grayscale = transforms.Grayscale()
transform = transforms.Compose([transforms.Lambda(lambda img: center_crop_square(img, min(*img.size))),
transforms.Resize((args.imsize, args.imsize)),
transforms.ToTensor(),
grayscale])
# Make probability distribution?
if args.prob:
transform.transforms.append(transforms.Lambda(lambda x: x / x.sum()))
# Dataset
channels = 1
if args.data == 'mnist':
# MNIST is already grayscale
transform.transforms.remove(grayscale)
train_data = datasets.MNIST('./', train=True, transform=transform, download=True)
test_data = datasets.MNIST('./', train=False, transform=transform, download=True)
elif args.data == 'kmnist':
# KMNIST is already grayscale
transform.transforms.remove(grayscale)
train_data = datasets.KMNIST('./', train=True, transform=transform, download=True)
test_data = datasets.KMNIST('./', train=False, transform=transform, download=True)
elif args.data == 'cifar10':
train_data = datasets.CIFAR10('./', train=True, transform=transform, download=True)
test_data = datasets.CIFAR10('./', train=False, transform=transform, download=True)
elif args.data == 'cifar100':
train_data = datasets.CIFAR100('./', train=True, transform=transform, download=True)
test_data = datasets.CIFAR100('./', train=False, transform=transform, download=True)
elif args.data == 'yale':
if not _check_exists('./CroppedYale'):
_download('http://vision.ucsd.edu/extyaleb/CroppedYaleBZip/CroppedYale.zip', 'yale_DB')
_unzip('./yale_DB.zip')
dataset = datasets.ImageFolder('CroppedYale', transform = transform)
train_data, test_data = torch.utils.data.random_split(dataset, lengths= [2000,452])
else:
raise Exception(f'unknown data {args.data}')
# Dataloader
train_loader = data.DataLoader(train_data, batch_size=args.bsz,
shuffle=shuffle, num_workers=4,
pin_memory=True, drop_last=droplast)
test_loader = data.DataLoader(test_data, batch_size=args.bsz,
shuffle=shuffle, num_workers=4,
pin_memory=True, drop_last=droplast)
return train_loader, test_loader, channels
def prepare_kmnist():
m = (0.5, )
st = (0.5, )
normalize = tf.Normalize(m, st)
train_set = datasets.KMNIST("../data",
download=True,
transform=tf.Compose(
[tf.ToTensor(), normalize]))
test_set = datasets.KMNIST("../data",
download=True,
train=False,
transform=tf.Compose([tf.ToTensor(),
normalize]))
return train_set, test_set
def __init__(self, mnist_type='MNIST', train=True, flatten=True):
if mnist_type == 'MNIST':
self.dataset = datasets.MNIST('../../data',
train=train,
download=True)
elif mnist_type == 'FashionMNIST':
self.dataset = datasets.FashionMNIST('../../data',
train=train,
download=True)
elif mnist_type == 'KMNIST':
self.dataset = datasets.KMNIST('../../data',
train=train,
download=True)
self.data = self.dataset.data
self.base_transform = transforms.ToTensor()
self.a_transform = transforms.Compose([
transforms.ToTensor(), # first, convert image to PyTorch tensor
transforms.ToPILImage()
])
self.b_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Lambda(add_mnist_noise),
transforms.Lambda(self.__threshold_func__)
])
self.targets = self.dataset.targets
self.OHs = OH_digits(self.targets.numpy().astype(int))
self.filtered_classes = []
self.filtered_nums = []
for i in range(10):
self.filtered_classes.append(self.data[self.targets == i])
self.filtered_nums.append(self.filtered_classes[i].shape[0])
self.flatten = flatten
def __init__(self, mnist_type='MNIST', train=True):
if mnist_type == 'MNIST':
self.dataset = datasets.MNIST('../../data',
train=train,
download=True)
elif mnist_type == 'FashionMNIST':
self.dataset = datasets.FashionMNIST('../../data',
train=train,
download=True)
elif mnist_type == 'KMNIST':
self.dataset = datasets.KMNIST('../../data',
train=train,
download=True)
self.data = self.dataset.data
self.base_transform = transforms.ToTensor()
self.a_transform = transforms.Compose([
transforms.ToTensor(), # first, convert image to PyTorch tensor
transforms.ToPILImage()
# transforms.Normalize((self.mean,), (self.std,)) # normalize inputs
])
self.b_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Lambda(lambda x: x + torch.rand(28, 28)),
transforms.Lambda(lambda x: self.__threshold_func__(x))
])
self.targets = self.dataset.targets
self.filtered_classes = []
self.filtered_nums = []
for i in range(10):
self.filtered_classes.append(self.data[self.targets == i])
self.filtered_nums.append(self.filtered_classes[i].shape[0])
def get_dataset(data_name, data_root, image_size, train):
transform = transforms.Compose(
[transforms.Resize(image_size),
transforms.ToTensor()])
if data_name == "mnist":
dataset = datasets.MNIST(root=data_root,
train=train,
transform=transform,
download=True)
elif data_name == "fushion-mnist":
dataset = datasets.FashionMNIST(root=data_root,
train=train,
transform=transform,
download=True)
elif data_name == "kmnist":
dataset = datasets.KMNIST(root=data_root,
train=train,
transform=transform,
download=True)
elif data_name == "emnist":
dataset = datasets.EMNIST(root=data_root,
split="byclass",
train=train,
transform=transform,
download=True)
else:
dataset = None
return dataset
def __init__(self, mnist_type='MNIST', train=True, fixed=False, flatten=True):
if mnist_type == 'MNIST':
self.dataset = datasets.MNIST('../../data', train=train, download=True)
elif mnist_type == 'FashionMNIST':
self.dataset = datasets.FashionMNIST('../../data', train=train, download=True)
elif mnist_type == 'KMNIST':
self.dataset = datasets.KMNIST('../../data', train=train, download=True)
self.data = self.dataset.data
self.mean = torch.mean(self.data.float())
self.std = torch.std(self.data.float())
self.transform = transforms.Compose([transforms.ToTensor()])
self.targets = self.dataset.targets
self.OHs = OH_digits(self.targets.numpy().astype(int))
self.fixed = fixed
self.filtered_classes = []
self.filtered_nums = []
for i in range(10):
self.filtered_classes.append(self.data[self.targets == i])
self.filtered_nums.append(self.filtered_classes[i].shape[0])
if fixed:
self.view_b_indices = []
for i in range(10):
self.view_b_indices.append(np.random.permutation(np.arange(len(self.data))[self.targets == i]))
self.flatten = flatten
def get_data(name, path="./data", train=True):
if name == "MNIST":
dataset = datasets.MNIST(path, train=train, download=True)
elif name == "KMNIST":
dataset = datasets.KMNIST(path, train=train, download=True)
images = dataset.data.float().unsqueeze(1)
return TensorDataset(images / 255., dataset.targets)
def prepare_cv_datasets_hyper(dataname, batch_size):
if dataname == 'mnist':
ordinary_train_dataset = dsets.MNIST(root='~/datasets/classification',
train=True,
transform=transforms.ToTensor(),
download=True)
elif dataname == 'kmnist':
ordinary_train_dataset = dsets.KMNIST(root='~/datasets/classification',
train=True,
transform=transforms.ToTensor(),
download=True)
elif dataname == 'fashion':
ordinary_train_dataset = dsets.FashionMNIST(
root='~/datasets/classification',
train=True,
transform=transforms.ToTensor(),
download=True)
elif dataname == 'cifar10':
train_transform = transforms.Compose([
transforms.ToTensor(
), # transforms.RandomHorizontalFlip(), transforms.RandomCrop(32,4),
transforms.Normalize((0.4914, 0.4822, 0.4465),
(0.247, 0.243, 0.261))
])
ordinary_train_dataset = dsets.CIFAR10(
root='~/datasets/classification',
train=True,
transform=train_transform,
download=True)
dataset_size = len(ordinary_train_dataset)
valid_proportion = 0.1
valid_size = int(np.floor(valid_proportion * dataset_size))
train_size = dataset_size - valid_size
trainingset, validationset = torch.utils.data.random_split(
ordinary_train_dataset, [train_size, valid_size])
train_loader = torch.utils.data.DataLoader(dataset=trainingset,
batch_size=batch_size,
shuffle=True,
drop_last=True,
num_workers=0)
valid_loader = torch.utils.data.DataLoader(dataset=validationset,
batch_size=batch_size,
shuffle=False,
drop_last=False,
num_workers=0)
train_eval_loader = torch.utils.data.DataLoader(dataset=trainingset,
batch_size=train_size,
shuffle=False,
drop_last=False,
num_workers=0)
num_classes = 10
return (train_eval_loader, train_loader, valid_loader, trainingset,
validationset, num_classes)
def get_kmnist_loaders(arguments, mean=(0.5,), std=(0.5,)):
print("Using mean", mean)
# (0.1918,), (0.3483,)
transformers = transforms.Compose([transforms.ToTensor(),
transforms.Normalize(mean, std)
])
train_set = datasets.KMNIST(
DATASET_PATH,
train=True,
download=True,
transform=transformers
)
test_set = datasets.KMNIST(
DATASET_PATH,
train=False,
download=True,
transform=transformers
)
return load(arguments, test_set, train_set)
def axi_loader(batch_size):
transform = transforms.Compose([
transforms.ToTensor(),
])
axi_data = datasets.KMNIST(root='./data',
train=True,
download=True,
transform=transform)
axi_loader = DataLoader(axi_data, batch_size=batch_size)
axi_x, _ = next(iter(axi_loader))
return axi_x
def get_dataset(data_name, data_root, stage, max_stage, train):
if data_name == "lsun":
transform = transforms.Compose([
transforms.Resize(4 * 2 ** min(stage, max_stage)),
transforms.CenterCrop(4 * 2 ** min(stage, max_stage)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
else:
transform = transforms.Compose([
transforms.Resize(4 * 2 ** min(stage, max_stage)),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5])])
if data_name == "mnist":
dataset = datasets.MNIST(root=data_root, train=train, transform=transform, download=True)
elif data_name == "fushion-mnist":
dataset = datasets.FashionMNIST(root=data_root, train=train, transform=transform, download=True)
elif data_name == "kmnist":
dataset = datasets.KMNIST(root=data_root, train=train, transform=transform, download=True)
elif data_name == "emnist":
dataset = datasets.EMNIST(root=data_root, split="balanced", train=train, transform=transform, download=True)
elif data_name == "cifar10":
dataset = datasets.CIFAR10(root=data_root, train=train, transform=transform, download=True)
elif data_name == "cifar100":
dataset = datasets.CIFAR100(root=data_root, train=train, transform=transform, download=True)
elif data_name == "lsun":
dataset = datasets.LSUN(root=data_root, classes="train", transform=transforms)
elif data_name == "stl10":
dataset = datasets.STL10(root=data_root, split="train", transform=transform, download=True)
else:
dataset = None
return dataset
def __init__(self, mnist_type='MNIST', train=True, fixed=False):
if mnist_type == 'MNIST':
self.dataset = datasets.MNIST('../../data',
train=train,
download=True)
elif mnist_type == 'FashionMNIST':
self.dataset = datasets.FashionMNIST('../../data',
train=train,
download=True)
elif mnist_type == 'KMNIST':
self.dataset = datasets.KMNIST('../../data',
train=train,
download=True)
self.data = self.dataset.data
self.mean = torch.mean(self.data.float())
self.std = torch.std(self.data.float())
self.transform = transforms.Compose([
transforms.ToTensor(
), # first, convert image to PyTorch tensor
# transforms.Lambda(lambda x: x/255.),
# transforms.Normalize((self.mean,), (self.std,)) # normalize inputs
])
self.targets = self.dataset.targets
self.fixed = fixed
self.filtered_classes = []
self.filtered_nums = []
for i in range(10):
self.filtered_classes.append(self.data[self.targets == i])
self.filtered_nums.append(self.filtered_classes[i].shape[0])
if fixed:
self.view_b_indices = []
for i in range(10):
self.view_b_indices.append(
np.random.permutation(
np.arange(len(self.data))[self.targets == i]))
def main():
# Training settings
parser = argparse.ArgumentParser(
description='Obtain confidence scores from trained networks')
parser.add_argument('--test-batch-size',
type=int,
default=200,
metavar='N',
help='input batch size for testing (default: 200)')
parser.add_argument('--no-cuda',
action='store_true',
default=False,
help='disables CUDA training')
parser.add_argument('--seed',
type=int,
default=1,
metavar='S',
help='random seed (default: 1)')
parser.add_argument('--arch',
type=str,
default='wrn',
metavar='ARC',
help='neural network arch')
parser.add_argument('--data',
type=str,
default='cifar10',
metavar='DT',
help='dataset the classifier was trained with')
parser.add_argument('--test-data',
type=str,
default='fakedata',
metavar='DT',
help='dataset used to test the classifier')
parser.add_argument('--lsun-path', type=str, help='path to LSUN dataset')
parser.add_argument('--save_path',
type=str,
default='data',
metavar='SP',
help='path to save the produced confidence')
parser.add_argument('--print-time',
type=bool,
default=False,
metavar='PT',
help='print elapsed time per batch')
parser.add_argument(
'--mode',
type=str,
default='msp',
metavar='M',
help=
'available modes: msp (maximum softmax probability), tmsp (t-softmax msp), tmsp0.5, tmsp5, tmsp10 (where number indicates nu value) odin. default: msp'
)
parser.add_argument('--T',
type=float,
default=1000.0,
metavar='T',
help='ODIN temperature scaling')
parser.add_argument('--eps',
type=float,
default=0.001,
metavar='EPS',
help='ODIN epsilon value')
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
torch.manual_seed(args.seed)
np.random.seed(args.seed)
if device.type == 'cuda':
torch.cuda.manual_seed(args.seed)
kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
mean, std = get_normal(args.data)
if args.test_data == "cifar10":
test_transform = trn.Compose(
[trn.ToTensor(), trn.Normalize(mean, std)])
test_loader = torch.utils.data.DataLoader(
datasets.CIFAR10('../data',
download=True,
train=False,
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "cifar10_bw":
test_transform = trn.Compose([
trn.Grayscale(),
trn.Resize((28, 28)),
trn.ToTensor(),
trn.Normalize(mean, std)
])
test_loader = torch.utils.data.DataLoader(
datasets.CIFAR10('../data',
download=True,
train=False,
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "svhn":
test_transform = trn.Compose(
[trn.ToTensor(), trn.Normalize(mean, std)])
test_loader = torch.utils.data.DataLoader(
datasets.SVHN('../data',
download=True,
split='test',
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "fakedata":
test_transform = trn.Compose(
[trn.ToTensor(), trn.Normalize(mean, std)])
test_loader = torch.utils.data.DataLoader(
datasets.FakeData(size=10000,
image_size=(3, 32, 32),
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "fakedata_bw":
test_transform = trn.Compose(
[trn.ToTensor(), trn.Normalize(mean, std)])
test_loader = torch.utils.data.DataLoader(
datasets.FakeData(size=10000,
image_size=(1, 28, 28),
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "fakedata_wm":
# wrong mean normalization
mean = (50, 50, 50)
test_transform = trn.Compose(
[trn.ToTensor(), trn.Normalize(mean, std)])
test_loader = torch.utils.data.DataLoader(
datasets.FakeData(size=10000,
image_size=(3, 32, 32),
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "fmnist":
test_transform = trn.Compose(
[trn.ToTensor(), trn.Normalize(mean, std)])
test_loader = torch.utils.data.DataLoader(
datasets.FashionMNIST('../data',
download=True,
train=False,
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "kmnist":
test_transform = trn.Compose(
[trn.ToTensor(), trn.Normalize(mean, std)])
test_loader = torch.utils.data.DataLoader(
datasets.KMNIST('../data',
download=True,
train=False,
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "mnist":
test_transform = trn.Compose([
trn.ToTensor(),
# lambda x : x.repeat(3,1,1) * torch.rand(3,1,1),
trn.Normalize(mean, std)
])
test_loader = torch.utils.data.DataLoader(
datasets.MNIST('../data',
download=True,
train=False,
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
elif args.test_data == "lsun":
test_transform = trn.Compose([
trn.Resize((32, 32)), #trn.CenterCrop(32),
trn.ToTensor(),
trn.Normalize(mean, std)
])
test_loader = torch.utils.data.DataLoader(
datasets.LSUN(args.lsun_path,
classes='test',
transform=test_transform),
batch_size=args.test_batch_size,
shuffle=True,
**kwargs)
if args.data == "cifar10":
Nc = 10
channels = 3
elif args.data == "svhn":
Nc = 10
channels = 3
elif args.data == "fmnist":
Nc = 10
channels = 1
elif args.data == "kmnist":
Nc = 10
channels = 1
if (args.mode == 'msp') | (args.mode == 'odin') | (args.mode == 'cb'):
nu = 0.0
elif args.mode == 'tmsp':
nu = 1.0
elif args.mode == 'tmsp0.5':
nu = 0.5
elif args.mode == 'tmsp5.0':
nu = 5.0
elif args.mode == 'tmsp10.0':
nu = 10.0
else:
print('mode not recognized!')
quit()
model_path = 'models'
model_path += '/' + args.data
model_name = args.arch + 'nu{}'.format(nu)
if args.arch == 'densenet':
densenet_depth = 100
model = DenseNet(densenet_depth, Nc, nu=nu).to(device)
elif args.arch == 'densenet_small':
densenet_depth = 10
model = DenseNet(densenet_depth, Nc, nu=nu).to(device)
elif args.arch == 'convnet':
model = ConvNet(Nc, channels=channels, nu=nu).to(device)
model.load_state_dict(
torch.load(model_path + '/' + model_name + '.pt', map_location=device))
score, is_hit = test(args, model, device, test_loader)
if not os.path.exists(args.save_path):
os.makedirs(args.save_path)
df = pd.DataFrame(data={'score': score, 'is_hit': is_hit})
df.to_csv('{}/{}_train{}_test{}_{}.csv'.format(args.save_path, args.arch,
args.data, args.test_data,
args.mode))
def main():
global args, best_err1, device, num_classes
args = get_args()
torch.manual_seed(args.random_seed)
# most cases have 10 classes
# if there are more, then it will be reassigned
num_classes = 10
best_err1 = 100
# define datasets
if args.target == "mnist":
transform_train = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
train_set_all = datasets.MNIST('data',
train=True,
transform=transform_train,
target_transform=None,
download=True)
# set aside 10000 examples from the training set for validation
train_set, val_set = torch.utils.data.random_split(
train_set_all, [50000, 10000])
# if we do experiments with variable target set size, this will take care of it
# by default the target set size is 50000
target_set_size = min(50000, args.target_set_size)
train_set, _ = torch.utils.data.random_split(
train_set, [target_set_size, 50000 - target_set_size])
test_set = datasets.MNIST('data',
train=False,
transform=transform_test,
target_transform=None,
download=True)
elif args.target == "kmnist":
transform_train = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
train_set_all = datasets.KMNIST('data',
train=True,
transform=transform_train,
target_transform=None,
download=True)
# set aside 10000 examples from the training set for validation
train_set, val_set = torch.utils.data.random_split(
train_set_all, [50000, 10000])
target_set_size = min(50000, args.target_set_size)
# if we do experiments with variable target set size, this will take care of it
train_set, _ = torch.utils.data.random_split(
train_set, [target_set_size, 50000 - target_set_size])
test_set = datasets.KMNIST('data',
train=False,
transform=transform_test,
target_transform=None,
download=True)
elif args.target == "k49":
num_classes = 49
train_images = np.load('./data/k49-train-imgs.npz')['arr_0']
test_images = np.load('./data/k49-test-imgs.npz')['arr_0']
train_labels = np.load('./data/k49-train-labels.npz')['arr_0']
test_labels = np.load('./data/k49-test-labels.npz')['arr_0']
transform_train = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
# set aside about 10% of training data for validation
train_set_all = K49Dataset(train_images,
train_labels,
transform=transform_train)
train_set, val_set = torch.utils.data.random_split(
train_set_all, [209128, 23237])
# currently we do not support variable target set size for k49
# enable this to use it
# target_set_size = min(209128, args.target_set_size)
# train_set, _ = torch.utils.data.random_split(
# train_set, [target_set_size, 209128 - target_set_size])
test_set = K49Dataset(test_images,
test_labels,
transform=transform_test)
elif args.target == "cifar10":
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_train = transforms.Compose(
[transforms.Resize((32, 32)),
transforms.ToTensor(), normalize])
transform_test = transforms.Compose(
[transforms.Resize((32, 32)),
transforms.ToTensor(), normalize])
train_set_all = datasets.CIFAR10('data',
train=True,
transform=transform_train,
target_transform=None,
download=True)
# set aside 5000 examples from the training set for validation
train_set, val_set = torch.utils.data.random_split(
train_set_all, [45000, 5000])
# if we do experiments with variable target set size, this will take care of it
target_set_size = min(45000, args.target_set_size)
train_set, _ = torch.utils.data.random_split(
train_set, [target_set_size, 45000 - target_set_size])
test_set = datasets.CIFAR10('data',
train=False,
transform=transform_test,
target_transform=None,
download=True)
elif args.target == "cifar100":
num_classes = 100
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_train = transforms.Compose(
[transforms.Resize((32, 32)),
transforms.ToTensor(), normalize])
transform_test = transforms.Compose(
[transforms.Resize((32, 32)),
transforms.ToTensor(), normalize])
train_set_all = datasets.CIFAR100('data',
train=True,
transform=transform_train,
target_transform=None,
download=True)
# set aside 5000 examples from the training set for validation
train_set, val_set = torch.utils.data.random_split(
train_set_all, [45000, 5000])
# if we do experiments with variable target set size, this will take care of it
target_set_size = min(45000, args.target_set_size)
train_set, _ = torch.utils.data.random_split(
train_set, [target_set_size, 45000 - target_set_size])
test_set = datasets.CIFAR100('data',
train=False,
transform=transform_test,
target_transform=None,
download=True)
# create data loaders
if args.baseline:
train_loader = torch.utils.data.DataLoader(
train_set, batch_size=args.num_base_examples, shuffle=True)
else:
train_loader = torch.utils.data.DataLoader(train_set,
batch_size=args.batch_size,
shuffle=True)
val_loader = torch.utils.data.DataLoader(val_set,
batch_size=args.batch_size,
shuffle=False)
test_loader = torch.utils.data.DataLoader(test_set,
batch_size=args.batch_size,
shuffle=False)
# create data loaders to get base examples
if args.source == "emnist":
train_set_source = datasets.EMNIST('data',
'letters',
train=True,
download=True,
transform=transform_train,
target_transform=None)
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
elif args.source == "mnist":
train_set_source = datasets.MNIST('data',
train=True,
download=True,
transform=transform_train,
target_transform=None)
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
elif args.source == "kmnist":
train_set_source = datasets.KMNIST('data',
train=True,
download=True,
transform=transform_train,
target_transform=None)
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
elif args.source == "cifar10":
train_set_source = datasets.CIFAR10('data',
train=True,
download=True,
transform=transform_train,
target_transform=None)
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
elif args.source == "cifar100":
train_set_source = datasets.CIFAR100('data',
train=True,
download=True,
transform=transform_train,
target_transform=None)
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
elif args.source == "svhn":
train_set_source = datasets.SVHN('data',
split='train',
download=True,
transform=transform_train,
target_transform=None)
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
elif args.source == "cub":
# modify the root depending on where you place the images
cub_data_root = './data/CUB_200_2011/images'
train_set_source = datasets.ImageFolder(cub_data_root,
transform=transform_train,
target_transform=None)
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
elif args.source == "fake":
# there is also an option to use random noise base examples
if args.target == "mnist":
num_channels = 1
dims = 28
else:
num_channels = 3
dims = 32
train_set_source = datasets.FakeData(size=5000,
image_size=(num_channels, dims,
dims),
num_classes=10,
transform=transform_train,
target_transform=None,
random_offset=0)
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
else:
# get the fixed images from the same dataset as the training data
train_set_source = train_set
train_loader_source = torch.utils.data.DataLoader(
train_set_source, batch_size=args.num_base_examples, shuffle=True)
if torch.cuda.is_available(): # checks whether a cuda gpu is available
device = torch.cuda.current_device()
print("use GPU", device)
print("GPU ID {}".format(torch.cuda.current_device()))
else:
print("use CPU")
device = torch.device('cpu') # sets the device to be CPU
train_loader_source_iter = iter(train_loader_source)
if args.balanced_source:
# use a balanced set of fixed examples - same number of examples per class
class_counts = {}
fixed_input = []
fixed_target = []
for batch_fixed_i, batch_fixed_t in train_loader_source_iter:
if sum(class_counts.values()) >= args.num_base_examples:
break
for fixed_i, fixed_t in zip(batch_fixed_i, batch_fixed_t):
if len(class_counts.keys()) < num_classes:
if int(fixed_t) in class_counts:
if class_counts[
int(fixed_t
)] < args.num_base_examples // num_classes:
class_counts[int(fixed_t)] += 1
fixed_input.append(fixed_i)
fixed_target.append(int(fixed_t))
else:
class_counts[int(int(fixed_t))] = 1
fixed_input.append(fixed_i)
fixed_target.append(int(fixed_t))
else:
if int(fixed_t) in class_counts:
if class_counts[
int(fixed_t
)] < args.num_base_examples // num_classes:
class_counts[int(fixed_t)] += 1
fixed_input.append(fixed_i)
fixed_target.append(int(fixed_t))
fixed_input = torch.stack(fixed_input).to(device=device)
fixed_target = torch.Tensor(fixed_target).to(device=device)
else:
# used for cross-dataset scenario - random selection of classes
# not taking into accound the original classes
fixed_input, fixed_target = next(train_loader_source_iter)
fixed_input = fixed_input.to(device=device)
fixed_target = fixed_target.to(device=device)
# define loss function (criterion)
criterion = nn.CrossEntropyLoss().to(device=device)
# start at uniform labels and then learn them
labels = torch.zeros((args.num_base_examples, num_classes),
requires_grad=True,
device=device)
labels = labels.new_tensor(
[[float(1.0 / num_classes) for e in range(num_classes)]
for i in range(args.num_base_examples)],
requires_grad=True,
device=device)
# define an optimizer for labels
labels_opt = torch.optim.Adam([labels])
# enable using meta-architectures for second-order meta-learning
# allows assigning fast weights
cudnn.benchmark = True
M.LeNetMeta.meta = True
M.AlexCifarNetMeta.meta = True
M.BasicBlockMeta.meta = True
M.BottleneckMeta.meta = True
M.ResNetMeta.meta = True
# define the models to use
if args.target == "cifar10" or args.target == "cifar100":
if args.resnet:
model = M.ResNetMeta(dataset=args.target,
depth=18,
num_classes=num_classes,
bottleneck=False,
device=device).to(device=device)
model_name = 'resnet'
else:
model = M.AlexCifarNetMeta(args).to(device=device)
model_name = 'alexnet'
else:
model = M.LeNetMeta(args).to(device=device)
model_name = 'LeNet'
optimizer = torch.optim.Adam(model.parameters())
if args.baseline:
create_json_experiment_log(fixed_target)
# remap the targets - only relevant in cross-dataset
fixed_target = remap_targets(fixed_target, num_classes)
# printing the labels helps ensure the seeds work
print('The labels of the fixed examples are')
print(fixed_target.tolist())
labels = one_hot(fixed_target.long(), num_classes)
# add smoothing to the baseline if selected
if args.label_smoothing > 0:
labels = create_smooth_labels(labels, args.label_smoothing,
num_classes)
# use the validation set to find a suitable number of iterations for training
num_baseline_steps, errors_list, num_steps_used = find_best_num_steps(
val_loader, criterion, fixed_input, labels)
print('Number of steps to use for the baseline: ' +
str(num_baseline_steps))
experiment_update_dict = {
'num_baseline_steps': num_baseline_steps,
'errors_list': errors_list,
'num_steps_used': num_steps_used
}
update_json_experiment_log_dict(experiment_update_dict)
if args.test_various_models:
assert args.target == "cifar10", "test various models is only meant to be used for CIFAR-10"
model_name_list = ['alexnet', 'LeNet', 'resnet']
for model_name_test in model_name_list:
# do 20 repetitions of training from scratch
for test_i in range(20):
print('Test repetition ' + str(test_i))
test_err1, test_loss = test(test_loader, model_name_test,
criterion, fixed_input, labels,
num_baseline_steps)
print('Test error (top-1 error):', test_err1)
experiment_update_dict = {
'test_top_1_error_' + model_name_test: test_err1,
'test_loss_' + model_name_test: test_loss,
'num_test_steps_' + model_name_test: num_baseline_steps
}
update_json_experiment_log_dict(experiment_update_dict)
else:
# do 20 repetitions of training from scratch
for test_i in range(20):
print('Test repetition ' + str(test_i))
test_err1, test_loss = test(test_loader, model_name, criterion,
fixed_input, labels,
num_baseline_steps)
print('Test error (top-1 error):', test_err1)
experiment_update_dict = {
'test_top_1_error': test_err1,
'test_loss': test_loss,
'num_test_steps': num_baseline_steps
}
update_json_experiment_log_dict(experiment_update_dict)
else:
create_json_experiment_log(fixed_target)
# start measuring time
start_time = time.time()
# initialize variables to decide when to restart a model
ma_list = []
ma_sum = 0
lowest_ma_sum = 999999999
current_num_steps = 0
num_steps_list = []
num_steps_from_min = 0
val_err1 = 100.0
val_loss = 5.0
num_steps_val = 0
with tqdm.tqdm(total=args.epochs) as pbar_epochs:
for epoch in range(0, args.epochs):
train_err1, train_loss, labels, model_loss, ma_list, ma_sum, lowest_ma_sum, current_num_steps, num_steps_list, num_steps_from_min, model, optimizer = \
train(train_loader, model, fixed_input, labels, criterion, labels_opt, epoch, optimizer,
ma_list, ma_sum, lowest_ma_sum, current_num_steps, num_steps_list, num_steps_from_min)
# evaluate on the validation set only every 5 epochs as it can be quite expensive to train a new model from scratch
if epoch % 5 == 4:
# calculate the number of steps to use
if len(num_steps_list) == 0:
num_steps_val = current_num_steps
else:
num_steps_val = int(np.mean(num_steps_list[-3:]))
val_err1, val_loss = validate(val_loader, model, criterion,
epoch, fixed_input, labels,
num_steps_val)
if val_err1 <= best_err1:
best_labels = labels.detach().clone()
best_num_steps = num_steps_val
best_err1 = min(val_err1, best_err1)
print('Current best val error (top-1 error):', best_err1)
pbar_epochs.update(1)
experiment_update_dict = {
'train_top_1_error': train_err1,
'train_loss': train_loss,
'val_top_1_error': val_err1,
'val_loss': val_loss,
'model_loss': model_loss,
'epoch': epoch,
'num_val_steps': num_steps_val
}
# save the best labels so that we can analyse them
if epoch == args.epochs - 1:
experiment_update_dict['labels'] = best_labels.tolist()
update_json_experiment_log_dict(experiment_update_dict)
print('Best val error (top-1 error):', best_err1)
# stop measuring time
experiment_update_dict = {'total_train_time': time.time() - start_time}
update_json_experiment_log_dict(experiment_update_dict)
# this does number of steps analysis - what happens if we do more or fewer steps for test training
if args.num_steps_analysis:
num_steps_add = [-50, -20, -10, 0, 10, 20, 50, 100]
for num_steps_add_item in num_steps_add:
# start measuring time for testing
start_time = time.time()
local_errs = []
local_losses = []
local_num_steps = best_num_steps + num_steps_add_item
print('Number of steps for training: ' + str(local_num_steps))
# each number of steps will have a robust estimate by using 20 repetitions
for test_i in range(20):
print('Test repetition ' + str(test_i))
test_err1, test_loss = test(test_loader, model_name,
criterion, fixed_input,
best_labels, local_num_steps)
local_errs.append(test_err1)
local_losses.append(test_loss)
print('Test error (top-1 error):', test_err1)
experiment_update_dict = {
'test_top_1_error': local_errs,
'test_loss': local_losses,
'total_test_time': time.time() - start_time,
'num_test_steps': local_num_steps
}
update_json_experiment_log_dict(experiment_update_dict)
else:
if args.test_various_models:
assert args.target == "cifar10", "test various models is only meant to be used for CIFAR-10"
model_name_list = ['alexnet', 'LeNet', 'resnet']
for model_name_test in model_name_list:
for test_i in range(20):
print(model_name_test)
print('Test repetition ' + str(test_i))
test_err1, test_loss = test(test_loader,
model_name_test, criterion,
fixed_input, best_labels,
best_num_steps)
print('Test error (top-1 error):', test_err1)
experiment_update_dict = {
'test_top_1_error_' + model_name_test: test_err1,
'test_loss_' + model_name_test: test_loss,
'total_test_time_' + model_name_test:
time.time() - start_time,
'num_test_steps_' + model_name_test: best_num_steps
}
update_json_experiment_log_dict(experiment_update_dict)
else:
for test_i in range(20):
print('Test repetition ' + str(test_i))
test_err1, test_loss = test(test_loader, model_name,
criterion, fixed_input,
best_labels, best_num_steps)
print('Test error (top-1 error):', test_err1)
experiment_update_dict = {
'test_top_1_error': test_err1,
'test_loss': test_loss,
'total_test_time': time.time() - start_time,
'num_test_steps': best_num_steps
}
update_json_experiment_log_dict(experiment_update_dict)
def kmnist():
return collect_download_configs(
lambda: datasets.KMNIST(ROOT, download=True), name="KMNIST")
def load_datasets(name, root, batch_size):
if name == "mnist":
train_dataset = datasets.MNIST(root=root,
download=True,
train=True,
transform=transforms.Compose([
transforms.Resize(28),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]))
train_dataloader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=8)
test_dataset = datasets.MNIST(root=root,
download=True,
train=False,
transform=transforms.Compose([
transforms.Resize(28),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]))
test_dataloader = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=8)
return train_dataloader, test_dataloader
elif name == "fmnist":
train_dataset = datasets.FashionMNIST(
root=root,
download=True,
train=True,
transform=transforms.Compose([
transforms.Resize(28),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]))
train_dataloader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=8)
test_dataset = datasets.FashionMNIST(root=root,
download=True,
train=False,
transform=transforms.Compose([
transforms.Resize(28),
transforms.ToTensor(),
transforms.Normalize([0.5],
[0.5]),
]))
test_dataloader = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=8)
return train_dataloader, test_dataloader
elif name == "kmnist":
train_dataset = datasets.KMNIST(root=root,
download=True,
train=True,
transform=transforms.Compose([
transforms.Resize(28),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]))
train_dataloader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=8)
test_dataset = datasets.KMNIST(root=root,
download=True,
train=False,
transform=transforms.Compose([
transforms.Resize(28),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]))
test_dataloader = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=8)
return train_dataloader, test_dataloader
elif name == "qmnist":
train_dataset = datasets.QMNIST(root=root,
download=True,
train=True,
transform=transforms.Compose([
transforms.Resize(28),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]))
train_dataloader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=8)
test_dataset = datasets.QMNIST(root=root,
download=True,
what="test50k",
train=False,
transform=transforms.Compose([
transforms.Resize(28),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]))
test_dataloader = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=8)
return train_dataloader, test_dataloader
elif name == "cifar10":
train_dataset = datasets.CIFAR10(root=root,
download=True,
train=True,
transform=transforms.Compose([
transforms.Resize(32),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(
(0.5, 0.5, 0.5),
(0.5, 0.5, 0.5)),
]))
train_dataloader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=8)
test_dataset = datasets.CIFAR10(root=root,
download=True,
train=False,
transform=transforms.Compose([
transforms.Resize(32),
transforms.ToTensor(),
transforms.Normalize(
(0.5, 0.5, 0.5),
(0.5, 0.5, 0.5)),
]))
test_dataloader = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=8)
return train_dataloader, test_dataloader
elif name == "cifar100":
train_dataset = datasets.CIFAR100(root=root,
download=True,
train=True,
transform=transforms.Compose([
transforms.Resize(32),
transforms.ToTensor(),
transforms.Normalize(
(0.5, 0.5, 0.5),
(0.5, 0.5, 0.5)),
]))
train_dataloader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=8)
test_dataset = datasets.CIFAR100(root=root,
download=True,
train=False,
transform=transforms.Compose([
transforms.Resize(32),
transforms.ToTensor(),
transforms.Normalize(
(0.5, 0.5, 0.5),
(0.5, 0.5, 0.5)),
]))
test_dataloader = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=8)
return train_dataloader, test_dataloader
def get_kmnist(path="./data", train=True):
dataset = datasets.KMNIST(path, train=train, download=True)
return ImageDataset(dataset.data.float().unsqueeze(1) / 255.,
dataset.targets, dataset.classes)
def build_datasets(self):
if self.if_use_loacl_file:
self.dataset_name = os.path.join(root, 'data', self.dataset_name)
assert os.path.exists(
self.dataset_name
), "There is no file named %s" % self.dataset_name
def is_valid(img_path):
"""验证图像是否符合要求"""
return (img_path.endswith('jpg') or img_path.endswith('jpeg')
or img_path.endswith('png')
) and os.path.getsize(img_path) > 10
if self.if_train_val:
if os.path.exists(os.path.join(
self.dataset_name, 'train')) and os.path.exists(
os.path.join(self.dataset_name, 'val')):
if not isinstance(self.transform, dict):
self.transform = {
x: self.transform
for x in ['train', 'val']
}
datasets = {
x: ImageFolder(os.path.join(self.dataset_name, x),
self.transform[x])
for x in ['train', 'val']
}
else:
all_datasets = ImageFolder(self.dataset_name,
transform=self.transform,
is_valid_file=is_valid)
train_size = int(self.hold_out_rate * len(all_datasets))
val_size = len(all_datasets) - train_size
train_dataset, val_dataset = random_split(
all_datasets, [train_size, val_size])
datasets = {'train': train_dataset, 'val': val_dataset}
dataloaders = {
x: DataLoader(datasets[x],
batch_size=opt.batch_size,
shuffle=opt.shuffle,
num_workers=opt.num_workers)
for x in ['train', 'val']
}
dataset_sizes = {x: len(datasets[x]) for x in ['train', 'val']}
try:
class_names = datasets['train'].classes
except:
class_names = all_datasets.classes
else:
datasets = ImageFolder(self.dataset_name,
transform=self.transform,
is_valid_file=is_valid)
dataloaders = DataLoader(datasets,
batch_size=opt.batch_size,
shuffle=opt.shuffle,
num_workers=opt.num_workers)
dataset_sizes = len(datasets)
class_names = datasets.classes
else:
assert self.dataset_name in [
'mnist', 'kmnist', 'cifa10'
], "Other datasets are not supported yet, you can load it local." # 其他的数据集暂时不支持
if not isinstance(self.transform, dict):
self.transform = {x: self.transform for x in ['train', 'val']}
if self.dataset_name == 'mnist':
train_dataset = td.MNIST(root='.',
train=True,
transform=self.transform['train'],
download=True)
val_dataset = td.MNIST(root='.',
train=False,
transform=self.transform['val'],
download=True)
elif self.dataset_name == 'kmnist':
train_dataset = td.KMNIST(root='.',
train=True,
transform=self.transform['train'],
download=True)
val_dataset = td.KMNIST(root='.',
train=False,
transform=self.transform['val'],
download=True)
else:
train_dataset = td.CIFAR10(root='.',
train=True,
transform=self.transform['train'],
download=True)
val_dataset = td.CIFAR10(root='.',
train=False,
transform=self.transform['val'],
download=True)
datasets = {'train': train_dataset, 'val': val_dataset}
dataloaders = {
x: DataLoader(datasets[x],
batch_size=opt.batch_size,
shuffle=opt.shuffle,
num_workers=opt.num_workers)
for x in ['train', 'val']
}
dataset_sizes = {x: len(datasets[x]) for x in ['train', 'val']}
class_names = datasets['train'].classes
class Result:
def __init__(self, ds, dl, dn, cn):
self.dataset = ds
self.dataloader = dl
self.dataset_size = dn
self.class_name = cn
res = Result(datasets, dataloaders, dataset_sizes, class_names)
return res
dt = layer.backward(dt)
def step(self, optimf):
for layer in self.layers:
layer.step(optimf)
# load data
numepoch = 100
batchsize = 128
import torch
from torchvision import datasets
from torchvision.transforms import ToTensor, Compose, Normalize
train_loader = torch.utils.data.DataLoader(datasets.KMNIST('../data',
train=True,
transform=Compose(
[ToTensor()])),
batch_size=batchsize,
shuffle=True)
test_loader = torch.utils.data.DataLoader(datasets.KMNIST('../data',
train=False,
transform=Compose([
ToTensor(),
])),
batch_size=batchsize,
shuffle=True)
net = Network()
criterion = SoftwaxCrossEntropy()
optimf = SGD()