Struct MnistDataset 

pub struct MnistDataset { /* private fields */ }

A loaded MNIST dataset stored entirely in memory.

Images are stored as Vec<u8> (28×28 = 784 bytes each). Labels are u8 values 0–9.

Implementations

impl MnistDataset

pub fn load( dir: impl AsRef<Path>, split: MnistSplit, ) -> Result<Self, MnistError>

Load MNIST from the given directory.

Expects the standard filenames (or .gz compressed versions):

• train-images-idx3-ubyte / train-images-idx3-ubyte.gz
• train-labels-idx1-ubyte / train-labels-idx1-ubyte.gz
• t10k-images-idx3-ubyte / t10k-images-idx3-ubyte.gz
• t10k-labels-idx1-ubyte / t10k-labels-idx1-ubyte.gz

pub fn from_raw( image_bytes: &[u8], label_bytes: &[u8], split: MnistSplit, ) -> Result<Self, MnistError>

Load from raw bytes (useful for embedded/testing).

pub fn synthetic(n: usize, split: MnistSplit) -> Self

Create a small synthetic MNIST-like dataset for testing.

Generates n random 28×28 images with random labels.

pub fn num_samples(&self) -> usize

Total number of samples.

pub fn image_dims(&self) -> (usize, usize)

Image dimensions: (rows, cols).

pub fn image_u8(&self, i: usize) -> &[u8]

Get the raw pixel values for sample i.

pub fn label(&self, i: usize) -> u8

Get the label for sample i.

pub fn split(&self) -> MnistSplit

Which split this dataset represents.

pub fn take(self, n: usize) -> Self

Take only the first n samples (useful for quick experiments).

Trait Implementations

impl Dataset for MnistDataset

fn len(&self) -> usize

Total number of samples in the dataset.

fn get(&self, index: usize) -> Sample

Retrieve the sample at position index.

fn feature_shape(&self) -> &[usize]

The shape of a single feature sample (without batch dim).

fn target_shape(&self) -> &[usize]

The shape of a single target sample (without batch dim).

fn name(&self) -> &str

Optional human-readable name.

fn is_empty(&self) -> bool

Whether the dataset is empty.

impl Debug for MnistDataset

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.