How to use the torchmeta.utils.data.CombinationMetaDataset function in torchmeta

import os
import json
import glob
import h5py
from PIL import Image, ImageOps

from torchmeta.utils.data import Dataset, ClassDataset, CombinationMetaDataset
from torchvision.datasets.utils import list_dir, download_url
from torchmeta.datasets.utils import get_asset


class Omniglot(CombinationMetaDataset):
    """
    The Omniglot dataset [1]. A dataset of 1623 handwritten characters from 
    50 different alphabets. 

    Parameters
    ----------
    root : string
        Root directory where the dataset folder `omniglot` exists.

    num_classes_per_task : int
        Number of classes per tasks. This corresponds to "N" in "N-way" 
        classification.

    meta_train : bool (default: `False`)
        Use the meta-train split of the dataset. If set to `True`, then the
        arguments `meta_val` and `meta_test` must be set to `False`. Exactly one 

import os
import json

from torchmeta.datasets.cifar100.base import CIFAR100ClassDataset
from torchmeta.datasets.utils import get_asset
from torchmeta.utils.data import ClassDataset, CombinationMetaDataset


class CIFARFS(CombinationMetaDataset):
    """
    The CIFAR-FS dataset, introduced in [1]. This dataset contains
    images of 100 different classes from the CIFAR100 dataset [2].

    Parameters
    ----------
    root : string
        Root directory where the dataset folder `cifar100` exists.

    num_classes_per_task : int
        Number of classes per tasks. This corresponds to `N` in `N-way` 
        classification.

    meta_train : bool (default: `False`)
        Use the meta-train split of the dataset. If set to `True`, then the
        arguments `meta_val` and `meta_test` must be set to `False`. Exactly one 

import numpy as np
from PIL import Image
import os
import io
import json
import glob
import h5py

from torchmeta.utils.data import Dataset, ClassDataset, CombinationMetaDataset
from torchvision.datasets.utils import download_url
from torchmeta.datasets.utils import get_asset


class CUB(CombinationMetaDataset):
    """
    The Caltech-UCSD Birds dataset, introduced in [1]. This dataset is based on
    images from 200 species of birds from the Caltech-UCSD Birds dataset [2].

    Parameters
    ----------
    root : string
        Root directory where the dataset folder `cub` exists.

    num_classes_per_task : int
        Number of classes per tasks. This corresponds to "N" in "N-way" 
        classification.

    meta_train : bool (default: `False`)
        Use the meta-train split of the dataset. If set to `True`, then the
        arguments `meta_val` and `meta_test` must be set to `False`. Exactly one 

import os
import pickle
from PIL import Image
import h5py
import json

from torchmeta.utils.data import Dataset, ClassDataset, CombinationMetaDataset
from torchvision.datasets.utils import download_file_from_google_drive


class MiniImagenet(CombinationMetaDataset):
    """
    The Mini-Imagenet dataset, introduced in [1]. This dataset contains images 
    of 100 different classes from the ILSVRC-12 dataset (Imagenet challenge). 
    The meta train/validation/test splits are taken from [2] for reproducibility.

    Parameters
    ----------
    root : string
        Root directory where the dataset folder `miniimagenet` exists.

    num_classes_per_task : int
        Number of classes per tasks. This corresponds to "N" in "N-way" 
        classification.

    meta_train : bool (default: `False`)
        Use the meta-train split of the dataset. If set to `True`, then the

import numpy as np
from PIL import Image
import h5py
import json
import os
import io
import pickle

from torchmeta.utils.data import Dataset, ClassDataset, CombinationMetaDataset
from torchvision.datasets.utils import download_file_from_google_drive


class TieredImagenet(CombinationMetaDataset):
    """
    The Tiered-Imagenet dataset, introduced in [1]. This dataset contains images 
    of 608 different classes from the ILSVRC-12 dataset (Imagenet challenge).

    Parameters
    ----------
    root : string
        Root directory where the dataset folder `tieredimagenet` exists.

    num_classes_per_task : int
        Number of classes per tasks. This corresponds to "N" in "N-way" 
        classification.

    meta_train : bool (default: `False`)
        Use the meta-train split of the dataset. If set to `True`, then the
        arguments `meta_val` and `meta_test` must be set to `False`. Exactly one 

import os
import json

from torchmeta.datasets.cifar100.base import CIFAR100ClassDataset
from torchmeta.datasets.utils import get_asset
from torchmeta.utils.data import ClassDataset, CombinationMetaDataset


class FC100(CombinationMetaDataset):
    """
    The Fewshot-CIFAR100 dataset, introduced in [1]. This dataset contains
    images of 100 different classes from the CIFAR100 dataset [2].

    Parameters
    ----------
    root : string
        Root directory where the dataset folder `cifar100` exists.

    num_classes_per_task : int
        Number of classes per tasks. This corresponds to `N` in `N-way` 
        classification.

    meta_train : bool (default: `False`)
        Use the meta-train split of the dataset. If set to `True`, then the
        arguments `meta_val` and `meta_test` must be set to `False`. Exactly one