2 releases

Uses old Rust 2015

0.0.2 Feb 27, 2018
0.0.1 Feb 27, 2018

#134 in Machine learning

MIT license

635 lines


This library provides access to common machine learning benchmarking datasets.

The library currently includes:

Things are currently very basic.


Each dataset can be downloaded and processed using a Builder class. The builder is customizable in each case.

extern crate vision;

use vision::mnist::{MNISTBuilder};

fn main() {
    let builder = MNISTBuilder::new();
    let mnist = builder.data_home("MNIST")
    println!("{}", mnist.train_imgs.len());

The MNIST object returned by the builder contains four public fields, train_imgs, train_labels, test_images and test_labels. The label fields are Vec<u8> types and the images are Vec<Vec<u8>>, each entry in the outermost Vec corresponds to a single datapoint.

Further preprocessing should be carried out by the user.


~123K SLoC