def run(self, step_limit):
self.train()
with tf.Session() as sess:
tf.global_variables_initializer().run()
dataset = svhn.SVHN()
train_data, train_label = dataset.get_trainset()
test_data, test_label = dataset.get_testset()
train_data = train_data.reshape(-1, 3072)
test_data = test_data.reshape(-1, 3072)
test_indices = np.arange(len(test_data))
np.random.shuffle(test_indices)
test_indices = test_indices[0:1000]
name = self.info()
path = "svhn/" + str(step_limit) + name
saver = NNutils.save(path, sess)
writer, writer_test, merged = NNutils.graph(path, sess)
step = sess.run(self.global_step)
while step < step_limit:
print("step :", step)
for start, end in zip(
range(0, len(train_data), self.batch_size),
range(self.batch_size, len(train_data),
self.batch_size)):
summary, \
_, loss, \
step = sess.run([merged,
self.training, self.cost,
self.global_step],
feed_dict={self.x: train_data[start:end],
self.y: train_label[start:end],
self.dropout_conv: 1.0,
self.dropout_normal: 1.0})
if step % 50 == 0:
writer.add_summary(summary, step)
print(step, datetime.now(), loss)
summary, \
loss, \
accuracy = sess.run([merged, self.cost, self.accuracy],
feed_dict={self.x: test_data,
self.y: test_label,
self.dropout_conv: 1.0,
self.dropout_normal: 1.0})
writer_test.add_summary(summary, step)
print("test results : ", accuracy, loss)
saver.save(sess, path + "/" + name + ".ckpt", step)
def select_dataset(name):
x_size, y_size, train_data, train_label, test_data, test_label = 0, 0, [], [] ,[] ,[] #초기화
if name == 'cifar':
dataset = cifar.CIFAR()
train_data, train_label, test_data, test_label = dataset.getdata()
train_data = train_data.reshape(-1, 3072)
test_data = test_data.reshape(-1, 3072)
x_size = 3072
y_size = 10
elif name == 'svhn':
dataset = svhn.SVHN()
train_data, train_label = dataset.get_trainset()
test_data, test_label = dataset.get_testset()
train_data = train_data.reshape(-1, 3072)
test_data = test_data.reshape(-1, 3072)
x_size = 3072
y_size = 10
elif name == 'mnist':
dataset = mnist.read_data_sets(flags.MNIST_DIR, one_hot=True)
train_data, train_label, test_data, test_label = dataset.train.images, dataset.train.labels, \
dataset.test.images, dataset.test.labels
x_size = 784
y_size = 10
elif name == 'news':
trainset = fetch_20newsgroups(data_home=flags.NEWS_DIR, subset='train')
testset = fetch_20newsgroups(data_home=flags.NEWS_DIR, subset='test')
vectorizer = TfidfVectorizer(analyzer='word', max_features=3072)
vectorizer.fit(trainset.data)
train_data = vectorizer.transform(trainset.data)
train_data = csr_matrix.todense(train_data)
train_label = trainset.target
train_label = NNutils.onehot(train_label, 20, list=True)
# print(train_label.shape)
test_data = vectorizer.transform(testset.data)
test_data = csr_matrix.todense(test_data)
test_label = testset.target
test_label = NNutils.onehot(test_label, 20, list=True)
x_size = 3072
y_size = 20
return Dataset(name, x_size, y_size, train_data, train_label, test_data,
test_label)
def test(nnName, dataName, CUDA_DEVICE, epsilon, temperature):
net1 = torch.load("../models/{}.pth".format(nnName))
optimizer1 = optim.SGD(net1.parameters(), lr = 0, momentum = 0)
net1.cuda(CUDA_DEVICE)
if dataName != "Uniform" and dataName != "Gaussian":
if dataName == "SVHN":
testsetout = svhn.SVHN("../data/SVHN", split='test', transform=transform, download=True)
elif dataName in ["HFlip","VFlip"]:
testsetout = torchvision.datasets.CIFAR10('../data', train=False, download=True,
transform=Flip[dataName])
elif dataName == "CelebA":
testsetout = torchvision.datasets.ImageFolder(
"../data/{}".format(dataName),
transform=transforms.Compose([transforms.CenterCrop(178), Resize(32), transform]))
else:
testsetout = torchvision.datasets.ImageFolder("../data/{}".format(dataName),
transform=transform)
testloaderOut = torch.utils.data.DataLoader(testsetout, batch_size=1,
shuffle=False, num_workers=2)
if nnName == "densenet10" or nnName == "wideresnet10":
testset = torchvision.datasets.CIFAR10(root='../data', train=False, download=True, transform=transform)
testloaderIn = torch.utils.data.DataLoader(testset, batch_size=1,
shuffle=False, num_workers=2)
if nnName == "densenet100" or nnName == "wideresnet100":
testset = torchvision.datasets.CIFAR100(root='../data', train=False, download=True, transform=transform)
testloaderIn = torch.utils.data.DataLoader(testset, batch_size=1,
shuffle=False, num_workers=2)
if dataName == "Gaussian":
d.testGaussian(net1, criterion, CUDA_DEVICE, testloaderIn, testloaderIn, nnName, dataName, epsilon, temperature)
m.metric(nnName, dataName)
elif dataName == "Uniform":
d.testUni(net1, criterion, CUDA_DEVICE, testloaderIn, testloaderIn, nnName, dataName, epsilon, temperature)
m.metric(nnName, dataName)
else:
d.testData(net1, criterion, CUDA_DEVICE, testloaderIn, testloaderOut, nnName, dataName, epsilon, temperature)
m.metric(nnName, dataName)
def collate(batch):
transposed = zip(*batch)
(images, input_labels, output_labels) = list(transposed)
images = torch.stack(images, 0)
input_labels = torch.LongTensor(input_labels)
output_labels = torch.LongTensor(output_labels)
return (images, input_labels, output_labels)
composed = transforms.Compose([
transforms.ToPILImage(),
transforms.Scale(100),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
train_dataset = svhn.SVHN('./svhn', split='train', transform=composed)
train_dataloader = DataLoader(train_dataset,
batch_size=BATCH_SIZE,
shuffle=True,
num_workers=0,
collate_fn=collate)
n_iters = len(train_dataloader)
val_dataset = svhn.SVHN('./svhn', split='val', transform=composed)
val_dataloader = DataLoader(val_dataset,
batch_size=1,
num_workers=0,
collate_fn=collate)
test_dataset = svhn.SVHN('./svhn', split='test', transform=composed)
test_dataloader = DataLoader(test_dataset,
