How to use learn2learn - 10 common examples

cuda=False,
        seed=42,
):
    random.seed(seed)
    np.random.seed(seed)
    th.manual_seed(seed)
    device = th.device('cpu')
    if cuda and th.cuda.device_count():
        th.cuda.manual_seed(seed)
        device = th.device('cuda')

    # Create Datasets
    train_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='train')
    valid_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='validation')
    test_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='test')
    train_dataset = l2l.data.MetaDataset(train_dataset)
    valid_dataset = l2l.data.MetaDataset(valid_dataset)
    test_dataset = l2l.data.MetaDataset(test_dataset)

    train_transforms = [
        l2l.data.transforms.NWays(train_dataset, ways),
        l2l.data.transforms.KShots(train_dataset, 2*shots),
        l2l.data.transforms.LoadData(train_dataset),
        l2l.data.transforms.RemapLabels(train_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    train_tasks = l2l.data.TaskDataset(train_dataset,
                                       task_transforms=train_transforms,
                                       num_tasks=20000)

    valid_transforms = [
        l2l.data.transforms.NWays(valid_dataset, ways),

def setUpClass(cls) -> None:
        cls.ds = TestDatasets()
        cls.meta_tensor_dataset = MetaDataset(cls.ds.tensor_dataset)
        cls.meta_str_dataset = MetaDataset(cls.ds.str_dataset)
        cls.meta_alpha_dataset = MetaDataset(cls.ds.alphabet_dataset)
        cls.mnist_dataset = MetaDataset(cls.ds.get_mnist())
        cls.omniglot_dataset = MetaDataset(cls.ds.get_omniglot())

test_transforms = [
        l2l.data.transforms.NWays(test_dataset, ways),
        l2l.data.transforms.KShots(test_dataset, 2*shots),
        l2l.data.transforms.LoadData(test_dataset),
        l2l.data.transforms.RemapLabels(test_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    test_tasks = l2l.data.TaskDataset(test_dataset,
                                      task_transforms=test_transforms,
                                      num_tasks=600)

    # Create model
    model = l2l.vision.models.MiniImagenetCNN(ways)
    model.to(device)
    maml = l2l.algorithms.MAML(model, lr=fast_lr, first_order=False)
    opt = optim.Adam(maml.parameters(), meta_lr)
    loss = nn.CrossEntropyLoss(size_average=True, reduction='mean')

    for iteration in range(num_iterations):
        opt.zero_grad()
        meta_train_error = 0.0
        meta_train_accuracy = 0.0
        meta_valid_error = 0.0
        meta_valid_accuracy = 0.0
        meta_test_error = 0.0
        meta_test_accuracy = 0.0
        for task in range(meta_batch_size):
            # Compute meta-training loss
            learner = maml.clone()
            batch = train_tasks.sample()
            evaluation_error, evaluation_accuracy = fast_adapt(batch,

download_location="/tmp/mnist"):
    transformations = transforms.Compose([
        transforms.ToTensor(),
        transforms.Normalize((0.1307,), (0.3081,)),
        lambda x: x.view(1, 1, 28, 28),
    ])

    mnist_train = l2l.data.MetaDataset(MNIST(download_location, train=True, download=True, transform=transformations))
    # mnist_test = MNIST(file_location, train=False, download=True, transform=transformations)

    train_gen = l2l.data.TaskGenerator(mnist_train, ways=ways, tasks=10000)
    # test_gen = l2l.data.TaskGenerator(mnist_test, ways=ways)

    model = Net(ways)
    model.to(device)
    meta_model = l2l.algorithms.MAML(model, lr=maml_lr)
    opt = optim.Adam(meta_model.parameters(), lr=lr)
    loss_func = nn.NLLLoss(reduction="sum")

    for iteration in range(iterations):
        iteration_error = 0.0
        iteration_acc = 0.0
        for _ in range(tps):
            learner = meta_model.clone()
            train_task = train_gen.sample()
            valid_task = train_gen.sample(task=train_task.sampled_task)

            # Fast Adaptation
            for step in range(fas):
                train_error, _ = compute_loss(train_task, device, learner, loss_func, batch=shots * ways)
                learner.adapt(train_error)

# Create Datasets
    train_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='train')
    valid_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='validation')
    test_dataset = l2l.vision.datasets.MiniImagenet(root='./data', mode='test')
    train_dataset = l2l.data.MetaDataset(train_dataset)
    valid_dataset = l2l.data.MetaDataset(valid_dataset)
    test_dataset = l2l.data.MetaDataset(test_dataset)

    train_transforms = [
        l2l.data.transforms.NWays(train_dataset, ways),
        l2l.data.transforms.KShots(train_dataset, 2*shots),
        l2l.data.transforms.LoadData(train_dataset),
        l2l.data.transforms.RemapLabels(train_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    train_tasks = l2l.data.TaskDataset(train_dataset,
                                       task_transforms=train_transforms,
                                       num_tasks=20000)

    valid_transforms = [
        l2l.data.transforms.NWays(valid_dataset, ways),
        l2l.data.transforms.KShots(valid_dataset, 2*shots),
        l2l.data.transforms.LoadData(valid_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
        l2l.data.transforms.RemapLabels(valid_dataset),
    ]
    valid_tasks = l2l.data.TaskDataset(valid_dataset,
                                       task_transforms=valid_transforms,
                                       num_tasks=600)

    test_transforms = [
        l2l.data.transforms.NWays(test_dataset, ways),

l2l.data.transforms.LoadData(train_dataset),
        l2l.data.transforms.RemapLabels(train_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    train_tasks = l2l.data.TaskDataset(train_dataset,
                                       task_transforms=train_transforms,
                                       num_tasks=20000)

    valid_transforms = [
        l2l.data.transforms.NWays(valid_dataset, ways),
        l2l.data.transforms.KShots(valid_dataset, 2*shots),
        l2l.data.transforms.LoadData(valid_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
        l2l.data.transforms.RemapLabels(valid_dataset),
    ]
    valid_tasks = l2l.data.TaskDataset(valid_dataset,
                                       task_transforms=valid_transforms,
                                       num_tasks=600)

    test_transforms = [
        l2l.data.transforms.NWays(test_dataset, ways),
        l2l.data.transforms.KShots(test_dataset, 2*shots),
        l2l.data.transforms.LoadData(test_dataset),
        l2l.data.transforms.RemapLabels(test_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    test_tasks = l2l.data.TaskDataset(test_dataset,
                                      task_transforms=test_transforms,
                                      num_tasks=600)

    # Create model
    model = l2l.vision.models.MiniImagenetCNN(ways)

l2l.data.transforms.LoadData(valid_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
        l2l.data.transforms.RemapLabels(valid_dataset),
    ]
    valid_tasks = l2l.data.TaskDataset(valid_dataset,
                                       task_transforms=valid_transforms,
                                       num_tasks=600)

    test_transforms = [
        l2l.data.transforms.NWays(test_dataset, ways),
        l2l.data.transforms.KShots(test_dataset, 2*shots),
        l2l.data.transforms.LoadData(test_dataset),
        l2l.data.transforms.RemapLabels(test_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    test_tasks = l2l.data.TaskDataset(test_dataset,
                                      task_transforms=test_transforms,
                                      num_tasks=600)

    # Create model
    model = l2l.vision.models.MiniImagenetCNN(ways)
    model.to(device)
    maml = l2l.algorithms.MAML(model, lr=fast_lr, first_order=False)
    opt = optim.Adam(maml.parameters(), meta_lr)
    loss = nn.CrossEntropyLoss(size_average=True, reduction='mean')

    for iteration in range(num_iterations):
        opt.zero_grad()
        meta_train_error = 0.0
        meta_train_accuracy = 0.0
        meta_valid_error = 0.0
        meta_valid_accuracy = 0.0

def main(lr=0.005, maml_lr=0.01, iterations=1000, ways=5, shots=1, tps=32, fas=5, device=torch.device("cpu"),
         download_location="/tmp/mnist"):
    transformations = transforms.Compose([
        transforms.ToTensor(),
        transforms.Normalize((0.1307,), (0.3081,)),
        lambda x: x.view(1, 1, 28, 28),
    ])

    mnist_train = l2l.data.MetaDataset(MNIST(download_location, train=True, download=True, transform=transformations))
    # mnist_test = MNIST(file_location, train=False, download=True, transform=transformations)

    train_gen = l2l.data.TaskGenerator(mnist_train, ways=ways, tasks=10000)
    # test_gen = l2l.data.TaskGenerator(mnist_test, ways=ways)

    model = Net(ways)
    model.to(device)
    meta_model = l2l.algorithms.MAML(model, lr=maml_lr)
    opt = optim.Adam(meta_model.parameters(), lr=lr)
    loss_func = nn.NLLLoss(reduction="sum")

    for iteration in range(iterations):
        iteration_error = 0.0
        iteration_acc = 0.0
        for _ in range(tps):
            learner = meta_model.clone()
            train_task = train_gen.sample()
            valid_task = train_gen.sample(task=train_task.sampled_task)

test_dataset = l2l.data.MetaDataset(test_dataset)

    train_transforms = [
        l2l.data.transforms.NWays(train_dataset, ways),
        l2l.data.transforms.KShots(train_dataset, 2*shots),
        l2l.data.transforms.LoadData(train_dataset),
        l2l.data.transforms.RemapLabels(train_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    train_tasks = l2l.data.TaskDataset(train_dataset,
                                       task_transforms=train_transforms,
                                       num_tasks=20000)

    valid_transforms = [
        l2l.data.transforms.NWays(valid_dataset, ways),
        l2l.data.transforms.KShots(valid_dataset, 2*shots),
        l2l.data.transforms.LoadData(valid_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
        l2l.data.transforms.RemapLabels(valid_dataset),
    ]
    valid_tasks = l2l.data.TaskDataset(valid_dataset,
                                       task_transforms=valid_transforms,
                                       num_tasks=600)

    test_transforms = [
        l2l.data.transforms.NWays(test_dataset, ways),
        l2l.data.transforms.KShots(test_dataset, 2*shots),
        l2l.data.transforms.LoadData(test_dataset),
        l2l.data.transforms.RemapLabels(test_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    test_tasks = l2l.data.TaskDataset(test_dataset,

num_tasks=20000)

    valid_transforms = [
        l2l.data.transforms.NWays(valid_dataset, ways),
        l2l.data.transforms.KShots(valid_dataset, 2*shots),
        l2l.data.transforms.LoadData(valid_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
        l2l.data.transforms.RemapLabels(valid_dataset),
    ]
    valid_tasks = l2l.data.TaskDataset(valid_dataset,
                                       task_transforms=valid_transforms,
                                       num_tasks=600)

    test_transforms = [
        l2l.data.transforms.NWays(test_dataset, ways),
        l2l.data.transforms.KShots(test_dataset, 2*shots),
        l2l.data.transforms.LoadData(test_dataset),
        l2l.data.transforms.RemapLabels(test_dataset),
        l2l.data.transforms.ConsecutiveLabels(train_dataset),
    ]
    test_tasks = l2l.data.TaskDataset(test_dataset,
                                      task_transforms=test_transforms,
                                      num_tasks=600)

    # Create model
    model = l2l.vision.models.MiniImagenetCNN(ways)
    model.to(device)
    maml = l2l.algorithms.MAML(model, lr=fast_lr, first_order=False)
    opt = optim.Adam(maml.parameters(), meta_lr)
    loss = nn.CrossEntropyLoss(size_average=True, reduction='mean')

    for iteration in range(num_iterations):