Skip to content

CIFAR

SplitCIFAR

Bases: Benchmark

SplitCIFAR benchmarks.

Source code in sequel/benchmarks/cifar.py 
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
class SplitCIFAR(Benchmark):
    """SplitCIFAR benchmarks."""

    @property
    def num_classes(self) -> int:
        return 100 if self.is_cifar_100 else 10

    @property
    def MEAN(self):
        if self.is_cifar_100:
            return CIFAR100_MEAN
        else:
            return CIFAR10_MEAN

    @property
    def STD(self):
        if self.is_cifar_100:
            return CIFAR100_STD
        else:
            return CIFAR10_STD

    @property
    def dimensions(self) -> List[int]:
        return [3, 32, 32]

    @property
    def num_classes(self):
        num_classes = 100 if self.is_cifar_100 else 10
        return num_classes

    @property
    def classes_per_task(self):
        assert self.num_classes % self.num_tasks == 0
        return self.num_classes // self.num_tasks

    def __init__(
        self,
        num_tasks: int,
        batch_size: int,
        fixed_class_order: Optional[List[int]] = None,
        is_cifar_100: bool = True,
        eval_batch_size: int = None,
        num_workers: int = 0,
        pin_memory: bool = True,
        subset: Optional[int] = None,
    ):
        """Inits the SplitCIFAR100/100 class. The `is_cifar100` boolean flag denotes which dataset is instantiated.

        Args:
            num_tasks (int): the number of tasks in the benchmark. Usually 20 for SplitCIFAR100 and 5 for SplitCIFAR10.
                Must be divisible by the number of classes.
            batch_size (int, optional): The train dataloader batch size. Defaults to 256.
            fixed_class_order (Optional[List[int]], optional): A list of integers denoting a custom fixed_class_order.
                If None, the alphabetical order is used. Defaults to None.
            is_cifar_100 (bool, optional): Boolean denoting whether SplitCIFAR100 or SplitCIFAR10 is selected. Defaults
                to True.
            eval_batch_size (int, optional): The validation dataloader batch size. If None, `eval_batch_size` is set to
                `batch_size`. Defaults to None.
            num_workers (int, optional): Dataloader number of workers. Defaults to 0.
            pin_memory (bool, optional): pin_memory argument for dataloaders. Defaults to True.
        """

        self.is_cifar_100 = is_cifar_100

        if fixed_class_order is None:
            fixed_class_order = list(range(self.num_classes))
        assert (
            torch.tensor(fixed_class_order).sort()[0] == torch.arange(0, self.num_classes)
        ).all(), "The fixed_class_order argument muct contain exactly once all integers from 0 to (num_classes-1)."

        self.fixed_class_order = fixed_class_order
        super().__init__(
            num_tasks=num_tasks,
            batch_size=batch_size,
            eval_batch_size=eval_batch_size,
            num_workers=num_workers,
            pin_memory=pin_memory,
            subset=subset,
        )

    def prepare_datasets(self) -> Tuple[ContinualDataset, ContinualDataset]:
        transform = T.Compose([T.ToTensor(), T.Normalize(self.MEAN, self.STD)])
        CIFAR_dataset = torchvision.datasets.CIFAR100 if self.is_cifar_100 else torchvision.datasets.CIFAR10
        self.cifar_train = CIFAR_dataset(DEFAULT_DATASET_DIR, train=True, download=True, transform=transform)
        self.cifar_test = CIFAR_dataset(DEFAULT_DATASET_DIR, train=False, download=True, transform=transform)

        self.trains, self.tests = {}, {}
        for t in range(1, self.num_tasks + 1):
            self.trains[t] = SplitDataset(t, self.classes_per_task, self.cifar_train, self.fixed_class_order)
            self.tests[t] = SplitDataset(t, self.classes_per_task, self.cifar_test, self.fixed_class_order)

        return self.trains, self.tests

    @classmethod
    def from_config(cls, config):
        # breakpoint()
        kwargs = cls.get_default_kwargs(config)
        kwargs["is_cifar_100"] = "100" in config.name
        kwargs["fixed_class_order"] = getattr(config, "fixed_class_order", None)
        return cls(**kwargs)

    def __repr__(self) -> str:
        return f"SplitCIFAR{self.num_classes}(num_tasks={self.num_tasks}, batch_size={self.batch_size})"

__init__(num_tasks, batch_size, fixed_class_order=None, is_cifar_100=True, eval_batch_size=None, num_workers=0, pin_memory=True, subset=None)

Inits the SplitCIFAR100/100 class. The is_cifar100 boolean flag denotes which dataset is instantiated.

Parameters:

Name Type Description Default
num_tasks int

the number of tasks in the benchmark. Usually 20 for SplitCIFAR100 and 5 for SplitCIFAR10. Must be divisible by the number of classes.

 required
batch_size int

The train dataloader batch size. Defaults to 256.

 required
fixed_class_order Optional[List[int]]

A list of integers denoting a custom fixed_class_order. If None, the alphabetical order is used. Defaults to None.

 None
is_cifar_100 bool

Boolean denoting whether SplitCIFAR100 or SplitCIFAR10 is selected. Defaults to True.

 True
eval_batch_size int

The validation dataloader batch size. If None, eval_batch_size is set to batch_size. Defaults to None.

 None
num_workers int

Dataloader number of workers. Defaults to 0.

 0
pin_memory bool

pin_memory argument for dataloaders. Defaults to True.

 True
Source code in sequel/benchmarks/cifar.py 
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
def __init__(
    self,
    num_tasks: int,
    batch_size: int,
    fixed_class_order: Optional[List[int]] = None,
    is_cifar_100: bool = True,
    eval_batch_size: int = None,
    num_workers: int = 0,
    pin_memory: bool = True,
    subset: Optional[int] = None,
):
    """Inits the SplitCIFAR100/100 class. The `is_cifar100` boolean flag denotes which dataset is instantiated.

    Args:
        num_tasks (int): the number of tasks in the benchmark. Usually 20 for SplitCIFAR100 and 5 for SplitCIFAR10.
            Must be divisible by the number of classes.
        batch_size (int, optional): The train dataloader batch size. Defaults to 256.
        fixed_class_order (Optional[List[int]], optional): A list of integers denoting a custom fixed_class_order.
            If None, the alphabetical order is used. Defaults to None.
        is_cifar_100 (bool, optional): Boolean denoting whether SplitCIFAR100 or SplitCIFAR10 is selected. Defaults
            to True.
        eval_batch_size (int, optional): The validation dataloader batch size. If None, `eval_batch_size` is set to
            `batch_size`. Defaults to None.
        num_workers (int, optional): Dataloader number of workers. Defaults to 0.
        pin_memory (bool, optional): pin_memory argument for dataloaders. Defaults to True.
    """

    self.is_cifar_100 = is_cifar_100

    if fixed_class_order is None:
        fixed_class_order = list(range(self.num_classes))
    assert (
        torch.tensor(fixed_class_order).sort()[0] == torch.arange(0, self.num_classes)
    ).all(), "The fixed_class_order argument muct contain exactly once all integers from 0 to (num_classes-1)."

    self.fixed_class_order = fixed_class_order
    super().__init__(
        num_tasks=num_tasks,
        batch_size=batch_size,
        eval_batch_size=eval_batch_size,
        num_workers=num_workers,
        pin_memory=pin_memory,
        subset=subset,
    )

SplitCIFAR10

Bases: SplitCIFAR

Source code in sequel/benchmarks/cifar.py 
120
121
122
123
124
125
126
127
128
class SplitCIFAR10(SplitCIFAR):
    def __init__(self, is_cifar_100=False, *args, **kwargs):
        """Helper class for SplitCIFAR10. Inherits from SplitCIFAR. Look at the parent class for a description of the
        class arguments.

        Args:
            is_cifar_100 (bool, optional): Set to False.
        """
        super().__init__(is_cifar_100=is_cifar_100, *args, **kwargs)

__init__(is_cifar_100=False, *args, **kwargs)

Helper class for SplitCIFAR10. Inherits from SplitCIFAR. Look at the parent class for a description of the class arguments.

Parameters:

Name Type Description Default
is_cifar_100 bool

Set to False.

 False
Source code in sequel/benchmarks/cifar.py 
121
122
123
124
125
126
127
128
def __init__(self, is_cifar_100=False, *args, **kwargs):
    """Helper class for SplitCIFAR10. Inherits from SplitCIFAR. Look at the parent class for a description of the
    class arguments.

    Args:
        is_cifar_100 (bool, optional): Set to False.
    """
    super().__init__(is_cifar_100=is_cifar_100, *args, **kwargs)

SplitCIFAR100

Bases: SplitCIFAR

Source code in sequel/benchmarks/cifar.py 
131
132
133
134
135
136
137
138
139
class SplitCIFAR100(SplitCIFAR):
    def __init__(self, is_cifar_100=True, *args, **kwargs):
        """Helper class for SplitCIFAR100. Inherits from SplitCIFAR. Look at the parent class for a description of the
        class arguments.

        Args:
            is_cifar_100 (bool, optional): Set to True.
        """
        super().__init__(is_cifar_100=is_cifar_100, *args, **kwargs)

__init__(is_cifar_100=True, *args, **kwargs)

Helper class for SplitCIFAR100. Inherits from SplitCIFAR. Look at the parent class for a description of the class arguments.

Parameters:

Name Type Description Default
is_cifar_100 bool

Set to True.

 True
Source code in sequel/benchmarks/cifar.py 
132
133
134
135
136
137
138
139
def __init__(self, is_cifar_100=True, *args, **kwargs):
    """Helper class for SplitCIFAR100. Inherits from SplitCIFAR. Look at the parent class for a description of the
    class arguments.

    Args:
        is_cifar_100 (bool, optional): Set to True.
    """
    super().__init__(is_cifar_100=is_cifar_100, *args, **kwargs)