def main(n, h, lr, epochs, bs, exp_dir):
# ---------------------------------------------------------------
# Setup
# ---------------------------------------------------------------
# Logging
log_dir = Path(exp_dir / 'logs/')
log_dir.mkdir(exist_ok=True)
logging.basicConfig(level=logging.DEBUG,
format='%(levelname)s-%(message)s',
handlers=[
logging.FileHandler(log_dir / 'experiment1'),
logging.StreamHandler()
])
# Configurations
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
logging.info(
'n={},h={},lr={},epochs={},batch size={},dest={},device={}'.format(
n, h, lr, epochs, bs, exp_dir, device))
torch.manual_seed(590238490)
# device = 'cpu'
train_loader = MNIST.train(bs)
test_loader = MNIST.test(bs)
model = VAE(h, n).to(device)
optimizer = optim.Adagrad(model.parameters(), lr=lr)
# ---------------------------------------------------------------
# Execute experiment
# ---------------------------------------------------------------
execute(model, optimizer, epochs, train_loader, test_loader, device, n,
exp_dir)
def main(n, h, epochs, bs, learning_rate):
# device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
train_loader = mnist_.train(bs)
test_loader = mnist_.test(bs)
model = VAE(h, n)
# model = model.to(device) # for Adagrad: https://github.com/pytorch/pytorch/issues/7321
optimizer = optim.Adam(model.parameters(), lr=learning_rate)
# optimizer = optim.Adagrad(model.parameters(), lr = learning_rate)
logging.info('n={},h={},lr={},epochs={},batch size={}'
.format(n, h, learning_rate, epochs, batch_size))
exp = MNIST_vae(model, optimizer, learning_rate, bs, n, train_loader, test_loader)
train_points, train_loss, test_points, test_loss, seq_train_loss, seq_test_loss = exp.execute(epochs)
return train_points, train_loss, test_points, test_loss, seq_train_loss, seq_test_loss
def main_discrete(base_dir=Path('../data/results')):
# ---------------------------------------------------------------
# Setup Logger
# ---------------------------------------------------------------
log_dir = Path(base_dir / 'comparison/')
log_dir.mkdir(exist_ok=True)
logging.basicConfig(level=logging.DEBUG,
format='%(levelname)s-%(message)s',
handlers=[
logging.FileHandler(log_dir / 'comparison'),
logging.StreamHandler()
])
# ---------------------------------------------------------------
# Common configurations
# ---------------------------------------------------------------
torch.manual_seed(590238490)
data = MNIST.test(64, dir='../data')
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
exp1(base_dir, data, device)
exp1_relu(base_dir, data, device)
exp3(base_dir, data, device)
exp4(base_dir, data, device)
exp6(base_dir, data, device)
exp7(base_dir, data, device)
def get_ordered_digits():
data = MNIST.test(64, dir='../data')
digits_tensors = []
digits_labels = []
found = []
for x, labels in data:
for i, digit in enumerate(labels):
if digit.item() not in found:
found.append(digit.item())
digits_tensors.append(x[i])
digits_labels.append(digit)
elif len(found) == 10:
break
if len(found) == 10:
break
torch_digits_tensors = torch.stack(digits_tensors)
torch_digits_labels = torch.stack(digits_labels)
id_sort = torch.argsort(torch.Tensor(digits_labels))
# select digits in ascending order
torch_digits_labels = torch_digits_labels[id_sort]
torch_digits_tensors = torch_digits_tensors[id_sort]
# Testing
# for i, data in enumerate(torch_digits_tensors):
# img = data.numpy().transpose(1, 2, 0)
# lbl = torch_digits_labels[i]
# print(lbl)
# cv2.imshow('', img)
# cv2.waitKey()
return torch_digits_tensors, torch_digits_labels
def get_dataloader(conf: str) -> Tuple[DataLoader, DataLoader]:
if conf.dataset.name == "mnist":
data = MNIST(conf)
return data.train_dataloader(), data.val_dataloader()
else:
raise Exception(f"Invalid dataset name: {conf.dataset.name}")
def get_data():
data = MNIST.test(4000, dir='../data')
return data