def test_validation_split(data_folder_teardown):
dir_path = os.path.dirname(os.path.realpath(__file__))
dir_path = os.path.join(dir_path, 'data')
trainloader, valloader, evalloader = get_mnist(root=dir_path,
train_labels=range(6),
eval_labels=range(6),
split_val=0)
size_full_test = len(evalloader.dataset)
trainloader, valloader, evalloader = get_mnist(root=dir_path,
train_labels=range(6),
eval_labels=range(6),
split_val=0.5)
assert size_full_test == len(valloader.dataset) + len(
evalloader.dataset)
def get_evalloader_unseen(arguments):
type_of_unseen = arguments['type_of_unseen']
dim_channels = arguments['dim_channels']
dim_input = arguments['dim_input']
split_labels = arguments['split_labels']
if type_of_unseen == 'random':
_, _, evalloader_unseen = get_random(number_of_channels=dim_channels,
img_dim=dim_input,
number_of_classes=10)
if type_of_unseen == 'unseen_classes':
_, _, evalloader_unseen = get_trainset(
train_labels=(),
eval_labels=test_labels,
)
if type_of_unseen == 'unseen_dataset':
res['unseen_dataset'] = unseen_evalset
assert trainset != unseen_evalset, 'Train Set must be different from Unseen Test Set'
transform = transforms.Compose([
transforms.Grayscale(num_output_channels=dim_channels),
transforms.Resize(dim_input),
transforms.ToTensor(),
])
if unseen_evalset == 'cifar10':
_, _, evalloader_unseen = get_cifar10(transform=transform)
if unseen_evalset == 'mnist':
_, _, evalloader_unseen = get_mnist(transform=transform)
if unseen_evalset == 'omniglot':
_, _, evalloader_unseen = get_omniglot(transform=transform,
download=False)
return evalloader_unseen
def test_number_of_train(data_folder_teardown):
dir_path = os.path.dirname(os.path.realpath(__file__))
dir_path = os.path.join(dir_path, 'data')
trainloader, evalloader, evalloader = get_mnist(root=dir_path,
split_train=5)
assert len(trainloader.dataset) == 5
def test_label_specification(data_folder_teardown):
dir_path = os.path.dirname(os.path.realpath(__file__))
dir_path = os.path.join(dir_path, 'data')
train_labels = np.random.randint(0, 10, 5)
eval_labels = np.random.randint(0, 10, 5)
trainloader, valloader, evalloader = get_mnist(
root=dir_path,
train_labels=train_labels,
split_val=0,
eval_labels=eval_labels)
assert np.isin(trainloader.dataset.targets, train_labels).all()
assert np.isin(evalloader.dataset.targets, eval_labels).all()
number_of_tests = args.number_of_tests
loss_type = args.loss_type
std_prior = args.std_prior
stds_prior = (std_prior, std_prior)
delta = args.delta
risks = np.linspace(0.01, 0.5, 50)
if torch.cuda.is_available():
device = 'cuda'
else:
device = 'cpu'
device = torch.device(device)
if trainset == 'mnist':
trainloader, valloader, evalloader = get_mnist(train_labels=range(10),
eval_labels=range(10),
batch_size=batch_size)
dim_input = 28
dim_channels = 1
if trainset == 'cifar10':
trainloader, evalloader = get_cifar10(batch_size=batch_size)
dim_input = 32
dim_channels = 3
seed_model = set_and_print_random_seed()
bay_net = GaussianClassifier(rho=rho,
stds_prior=stds_prior,
dim_input=dim_input,
number_of_classes=10,
dim_channels=dim_channels)
bay_net.to(device)
_, _, evalloader_unseen = get_random(number_of_channels=dim_channels, img_dim=dim_input, number_of_classes=10)
if type_of_unseen == 'unseen_classes':
_, _, evalloader_unseen = get_trainset(train_labels=(), eval_labels=range(split_labels, 10, ), )
res['split_labels'] = split_labels
if type_of_unseen == 'unseen_dataset':
res['unseen_dataset'] = unseen_evalset
assert trainset != unseen_evalset, 'Train Set must be different from Unseen Test Set'
transform = transforms.Compose([
transforms.Grayscale(num_output_channels=dim_channels),
transforms.Resize(dim_input),
transforms.ToTensor(),
])
if unseen_evalset == 'cifar10':
_, _, evalloader_unseen = get_cifar10(transform=transform)
if unseen_evalset == 'mnist':
_, _, evalloader_unseen = get_mnist(transform=transform)
if unseen_evalset == 'omniglot':
_, _, evalloader_unseen = get_omniglot(transform=transform, download=False)
seed_model = set_and_print_random_seed()
bay_net = GaussianClassifier(rho=rho, stds_prior=stds_prior, dim_input=dim_input, number_of_classes=10, dim_channels=dim_channels)
bay_net.to(device)
criterion = CrossEntropyLoss()
if loss_type == 'uniform':
step_function = uniform
loss = BBBLoss(bay_net, criterion, step_function)
elif loss_type == 'exp':
def step_function(batch_idx, number_of_batches):
return 2 ** (number_of_batches - batch_idx) / (2 ** number_of_batches - 1)
##### TO CHANGE ######
nb_of_runs = args.nb_of_runs
nb_of_epochs = args.nb_of_epochs
nb_of_tests = args.nb_of_tests
rho = args.rho
######################
if torch.cuda.is_available():
device = "cuda"
else:
device = "cpu"
device = torch.device(device)
print(device)
trainloader, valloader, evalloader = get_mnist(batch_size=32)
criterion = nn.CrossEntropyLoss()
def train_bayesian_modular_with_one_different(
model,
optimizer,
loss,
observables,
number_of_epochs,
trainloader,
valloader=None,
number_of_tests=10,
output_dir_tensorboard=None,
output_dir_results=None,
device='cpu',