torchtraining.functional.data module

torchtraining.functional.data.random_split(dataset: torch.utils.data.dataset.Dataset, *p: float, generator=<torch._C.Generator object>)

Randomly split a dataset into non-overlapping new datasets of given proportions.

Works like torch.utils.data.random_split except data is splitted on [0, 1] proportions instead of length.

Example:

train, validation = tt.functional.data.random_split(
    torchvision.datasets.CIFAR10(
        root=".",
        download=True
        transform=torchvision.transforms.ToTensor(),
    ),
    0.8,
    0.2,
)

Above would be split dataset into 80% train and 20% validation.

Parameters

• dataset (torch.utils.data.Dataset) – Dataset to be split

• *p (float) – Floating point values in the [0, 1]. All of them should sum to 1 (if not they will be normalized to [0, 1] range). Split dataset according to those proportions.

• generator (Generator) – Generator used for the random permutation.

Returns

Tuple containing splitted datasets.

Return type

Tuple[torch.utils.data.Dataset]