dataset_mnist03_h5

Classes

Mnist03Dataset

Dataset class for MNIST03 (MNIST numbers 0-3) stored in HDF5 format.

Module Contents

class dataset_mnist03_h5.Mnist03Dataset(h5_path, split='train', normalize=True, flatten=True)

Dataset class for MNIST03 (MNIST numbers 0-3) stored in HDF5 format. Split in train, val and test set, image data and corresponding labels for each split are accessible through

/<split>/images : uint8 [N, 28, 28] /<split>/labels : int64 [N]

Args:
h5_path: Path to HDF5 file split: Dataset split (“train”, “val” or “test”). Default is “train”. normalize: bool

If True, normalize the images to [0, 1]. Default is True.

flatten: bool
If True, flatten the images from (1, 28, 28) to (784,). Default is True.

h5_path = ''

split = 'train'

normalize = True

flatten = True

_file = None

_pid = None

_length = None

_ensure_open()

Open the HDF5 file

Returns: h5py.File: A handle to the open HDF5 file (read-only)

__len__(): Number of samples in the current split.

__getitem__(idx)

Get a sample from the data set Args:

idx: index of sample

Returns:

X_t: torch.Tensor: sample image tensor
y_t: torch.Tensor: sample label tensor

__del__(): Close the HDF5 file