def main(args):
path = pathlib.Path('./src/gkernel')
if not path.is_file():
subprocess.call(["make"], cwd="./src", shell=True)
dataset = TUDataset(root=f'{args.dir}/Pytorch_geometric/{args.dataset}', name=args.dataset)
if dataset.num_features == 0:
max_degree = -1
for data in dataset:
edge_index = data.edge_index
degrees = Counter(list(map(int, edge_index[0])))
if max_degree < max(degrees.values()):
max_degree = max(degrees.values())
dataset.transform = OneHotDegree(max_degree=max_degree, cat=False)
path = pathlib.Path(f'{args.dir}/GraphML/{args.dataset}/{args.dataset.lower()}_{args.kernel}.kernel')
if not path.is_file():
save_to_graphml(dataset, f'{args.dir}/GraphML/{args.dataset}')
cmd = ['./src/gkernel']
cmd.append('-k')
cmd.append(args.kernel)
if args.parameter:
cmd.append('-p')
cmd.append(args.parameter)
cmd.append('-i')
cmd.append(f'{args.dir}/GraphML/{args.dataset}/{args.dataset.lower()}.list')
cmd.append('-g')
cmd.append(f'{args.dir}/GraphML/{args.dataset}/data/')
cmd.append('-o')
cmd.append(f'{args.dir}/GraphML/{args.dataset}/{args.dataset.lower()}_{args.kernel}.kernel')
subprocess.call(cmd)
K = read_kernel_matrix(f'{args.dir}/GraphML/{args.dataset}/{args.dataset.lower()}_{args.kernel}.kernel')
y = dataset.data.y.data.numpy()
ev = Evaluation(K, y, args, verbose=True)
accs = ev.evaluate(dataset)
outputWidth=128,
outputHeight=128)
options['data']['dir'] = options["globals"][dataset.value]
datasetHC, datasetPC = get_datasets(options, dataset=dataset)
config = get_config(trainer=ConstrainedAAE,
options=options,
optimizer='ADAM',
intermediateResolutions=[16, 16],
dropout_rate=0.1,
dataset=datasetHC)
config.kappa = 1.0
config.scale = 10.0
config.rho = 1.0
# Create an instance of the model and train it
model = ConstrainedAAE(tf.Session(),
config,
network=constrained_adversarial_autoencoder_Chen)
# Train it
model.train(datasetHC)
# Evaluate
Evaluation.evaluate(
datasetPC,
model,
options,
description=f"{type(datasetHC).__name__}-{options['threshold']}",
epoch=str(options['train']['numEpochs']))
def main(model_name, model_path):
conf = {}
conf['z_score'] = True
# Setup: initialize the hyperparameters/variables
num_epochs = 5 # Number of full passes through the dataset
batch_size = 128 # Number of samples in each minibatch
learning_rate = 1e-5
seed = np.random.seed(
1) # Seed the random number generator for reproducibility
p_val = 0.1 # Percent of the overall dataset to reserve for validation
p_test = 0.2 # Percent of the overall dataset to reserve for testing
val_every_n = 100 #
early_stop_counter = 0
early_stop_max = 7
is_converged = False
# TODO: Convert to Tensor - you can later add other transformations, such as Scaling here
transform = transforms.Compose(
[transforms.Resize(512), transforms.ToTensor()])
# Check if your system supports CUDA
use_cuda = torch.cuda.is_available()
# Setup GPU optimization if CUDA is supported
if use_cuda:
computing_device = torch.device("cuda")
extras = {"num_workers": 0, "pin_memory": True}
print("CUDA is supported")
else: # Otherwise, train on the CPU
computing_device = torch.device("cpu")
extras = False
print("CUDA NOT supported")
train_loader, val_loader, test_loader, _ = create_balanced_split_loaders(
batch_size,
seed,
transform=transform,
p_val=p_val,
p_test=p_test,
shuffle=True,
show_sample=False,
extras=extras,
z_score=conf['z_score'])
if model_name == 'intensive':
model = IntensiveCNN()
model = model.to(computing_device)
model.load_state_dict(torch.load(model_path)['model_state_dict'])
elif model_name == 'baseline':
model = BasicCNN()
model = model.to(computing_device)
model.load_state_dict(torch.load(model_path))
model.eval()
labels_all = []
predictions_all = []
model.eval()
with torch.no_grad():
for data in test_loader:
images, labels = data
images, labels = images.to(computing_device), labels.to(
computing_device)
labels_all.append(labels)
output = model(images)
predictions = output > 0.5
predictions_all.append(predictions)
labels = torch.cat(labels_all, 0)
predctions = torch.cat(predictions_all, 0)
eval = Evaluation(predctions.float(), labels)
eval.evaluate()