clarena.cl_datasets.permuted_cifar10

  1r"""
  2The submodule in `cl_datasets` for Permuted CIFAR-10 dataset.
  3"""
  4
  5__all__ = ["PermutedCIFAR10"]
  6
  7import logging
  8from typing import Callable
  9
 10import torch
 11from torch.utils.data import Dataset, random_split
 12from torchvision.datasets import CIFAR10
 13from torchvision.transforms import transforms
 14
 15from clarena.cl_datasets import CLPermutedDataset
 16
 17# always get logger for built-in logging in each module
 18pylogger = logging.getLogger(__name__)
 19
 20
 21class PermutedCIFAR10(CLPermutedDataset):
 22    r"""Permuted CIFAR-10 dataset. The [CIFAR-10 dataset](https://www.cs.toronto.edu/~kriz/cifar.html) is a subset of the [80 million tiny images dataset](https://people.csail.mit.edu/torralba/tinyimages/). It consists of 50,000 training and 10,000 test images of 10 classes, each 32x32 color image."""
 23
 24    original_dataset_python_class: type[Dataset] = CIFAR10
 25    r"""The original dataset class."""
 26
 27    def __init__(
 28        self,
 29        root: str,
 30        num_tasks: int,
 31        validation_percentage: float,
 32        batch_size: int | dict[int, int] = 1,
 33        num_workers: int | dict[int, int] = 0,
 34        custom_transforms: (
 35            Callable
 36            | transforms.Compose
 37            | None
 38            | dict[int, Callable | transforms.Compose | None]
 39        ) = None,
 40        repeat_channels: int | None | dict[int, int | None] = None,
 41        to_tensor: bool | dict[int, bool] = True,
 42        resize: tuple[int, int] | None | dict[int, tuple[int, int] | None] = None,
 43        permutation_mode: str = "first_channel_only",
 44        permutation_seeds: dict[int, int] | None = None,
 45    ) -> None:
 46        r"""
 47        **Args:**
 48        - **root** (`str`): the root directory where the original CIFAR-10 data 'cifar-10-python/' live.
 49        - **num_tasks** (`int`): the maximum number of tasks supported by the CL dataset. This decides the valid task IDs from 1 to `num_tasks`.
 50        - **validation_percentage** (`float`): the percentage to randomly split some training data into validation data.
 51        - **batch_size** (`int` | `dict[int, int]`): the batch size for train, val, and test dataloaders.
 52        If it is a dict, the keys are task IDs and the values are the batch sizes for each task. If it is an `int`, it is the same batch size for all tasks.
 53        - **num_workers** (`int` | `dict[int, int]`): the number of workers for dataloaders.
 54        If it is a dict, the keys are task IDs and the values are the number of workers for each task. If it is an `int`, it is the same number of workers for all tasks.
 55        - **custom_transforms** (`transform` or `transforms.Compose` or `None` or dict of them): the custom transforms to apply ONLY to the TRAIN dataset. Can be a single transform, composed transforms, or no transform. `ToTensor()`, normalization, permute, and so on are not included.
 56        If it is a dict, the keys are task IDs and the values are the custom transforms for each task. If it is a single transform or composed transforms, it is applied to all tasks. If it is `None`, no custom transforms are applied.
 57        - **repeat_channels** (`int` | `None` | dict of them): the number of channels to repeat for each task. Default is `None`, which means no repeat.
 58        If it is a dict, the keys are task IDs and the values are the number of channels to repeat for each task. If it is an `int`, it is the same number of channels to repeat for all tasks. If it is `None`, no repeat is applied.
 59        - **to_tensor** (`bool` | `dict[int, bool]`): whether to include the `ToTensor()` transform. Default is `True`.
 60        If it is a dict, the keys are task IDs and the values are whether to include the `ToTensor()` transform for each task. If it is a single boolean value, it is applied to all tasks.
 61        - **resize** (`tuple[int, int]` | `None` or dict of them): the size to resize the images to. Default is `None`, which means no resize.
 62        If it is a dict, the keys are task IDs and the values are the sizes to resize for each task. If it is a single tuple of two integers, it is applied to all tasks. If it is `None`, no resize is applied.
 63        - **permutation_mode** (`str`): the mode of permutation; one of:
 64            1. 'all': permute all pixels.
 65            2. 'by_channel': permute channel by channel separately. All channels are applied the same permutation order.
 66            3. 'first_channel_only': permute only the first channel.
 67        - **permutation_seeds** (`dict[int, int]` | `None`): the dict of seeds for permutation operations used to construct each task. Keys are task IDs and the values are permutation seeds for each task. Default is `None`, which creates a dict of seeds from 0 to `num_tasks`-1.
 68        """
 69        super().__init__(
 70            root=root,
 71            num_tasks=num_tasks,
 72            batch_size=batch_size,
 73            num_workers=num_workers,
 74            custom_transforms=custom_transforms,
 75            repeat_channels=repeat_channels,
 76            to_tensor=to_tensor,
 77            resize=resize,
 78            permutation_mode=permutation_mode,
 79            permutation_seeds=permutation_seeds,
 80        )
 81
 82        self.validation_percentage: float = validation_percentage
 83        r"""The percentage to randomly split some training data into validation data."""
 84
 85    def prepare_data(self) -> None:
 86        r"""Download the original CIFAR dataset if haven't."""
 87
 88        if self.task_id != 1:
 89            return  # download all original datasets only at the beginning of first task
 90
 91        CIFAR10(root=self.root_t, train=True, download=True)
 92        CIFAR10(root=self.root_t, train=False, download=True)
 93
 94        pylogger.debug(
 95            "The original CIFAR dataset has been downloaded to %s.", self.root_t
 96        )
 97
 98    def train_and_val_dataset(self) -> tuple[Dataset, Dataset]:
 99        """Get the training and validation dataset of task `self.task_id`.
100
101        **Returns:**
102        - **train_and_val_dataset** (`tuple[Dataset, Dataset]`): the train and validation dataset of task `self.task_id`.
103        """
104        dataset_train_and_val = CIFAR10(
105            root=self.root_t,
106            train=True,
107            transform=self.train_and_val_transforms(),
108            target_transform=self.target_transform(),
109            download=False,
110        )
111
112        return random_split(
113            dataset_train_and_val,
114            lengths=[1 - self.validation_percentage, self.validation_percentage],
115            generator=torch.Generator().manual_seed(
116                42
117            ),  # this must be set fixed to make sure the datasets across experiments are the same. Don't handle it to global seed as it might vary across experiments
118        )
119
120    def test_dataset(self) -> Dataset:
121        r"""Get the test dataset of task `self.task_id`.
122
123        **Returns:**
124        - **test_dataset** (`Dataset`): the test dataset of task `self.task_id`.
125        """
126        dataset_test = CIFAR10(
127            root=self.root_t,
128            train=False,
129            transform=self.test_transforms(),
130            target_transform=self.target_transform(),
131            download=False,
132        )
133
134        return dataset_test