Class Base

core/attacks/base.py:44–409

Base class for backdoor training and testing. Args: train_dataset (types in support_list): Benign training dataset. test_dataset (types in support_list): Benign testing dataset. model (torch.nn.Module): Network. loss (torch.nn.Module): Loss. schedule (dic

class Base(object):
    """Base class for backdoor training and testing.

    Args:
        train_dataset (types in support_list): Benign training dataset.
        test_dataset (types in support_list): Benign testing dataset.
        model (torch.nn.Module): Network.
        loss (torch.nn.Module): Loss.
        schedule (dict): Training or testing global schedule. Default: None.
        seed (int): Global seed for random numbers. Default: 0.
        deterministic (bool): Sets whether PyTorch operations must use "deterministic" algorithms.
            That is, algorithms which, given the same input, and when run on the same software and hardware,
            always produce the same output. When enabled, operations will use deterministic algorithms when available,
            and if only nondeterministic algorithms are available they will throw a RuntimeError when called. Default: False.
    """

    def __init__(self, train_dataset, test_dataset, model, loss, schedule=None, seed=0, deterministic=False):
        assert isinstance(train_dataset, support_list), 'train_dataset is an unsupported dataset type, train_dataset should be a subclass of our support list.'
        self.train_dataset = train_dataset

        assert isinstance(test_dataset, support_list), 'test_dataset is an unsupported dataset type, test_dataset should be a subclass of our support list.'
        self.test_dataset = test_dataset
        self.model = model
        self.loss = loss
        self.global_schedule = deepcopy(schedule)
        self.current_schedule = None
        self._set_seed(seed, deterministic)

    def _set_seed(self, seed, deterministic):
        # Use torch.manual_seed() to seed the RNG for all devices (both CPU and CUDA).
        torch.manual_seed(seed)

        # Set python seed
        random.seed(seed)

        # Set numpy seed (However, some applications and libraries may use NumPy Random Generator objects,
        # not the global RNG (https://numpy.org/doc/stable/reference/random/generator.html), and those will
        # need to be seeded consistently as well.)
        np.random.seed(seed)

        os.environ['PYTHONHASHSEED'] = str(seed)

        if deterministic:
            torch.backends.cudnn.benchmark = False
            torch.use_deterministic_algorithms(True)
            # torch.use_deterministic_algorithms(True, warn_only=True)
            torch.backends.cudnn.deterministic = True
            os.environ['CUBLAS_WORKSPACE_CONFIG'] = ':4096:8'
            # Hint: In some versions of CUDA, RNNs and LSTM networks may have non-deterministic behavior.
            # If you want to set them deterministic, see torch.nn.RNN() and torch.nn.LSTM() for details and workarounds.

    def _seed_worker(self, worker_id):
        worker_seed = torch.initial_seed() % 2**32
        np.random.seed(worker_seed)
        random.seed(worker_seed)

    def get_model(self):
        return self.model