Skip to contents

MNIST Digits Dataset

Source: R/data.R
digits.Rd

MNIST handwritten digit dataset containing grayscale images of digits 0-9. Each image is 28x28 pixels flattened to 784 features.

Usage

digits

Format

A matrix (dense) with pixel intensities. Rows are samples, columns are features.

Source

MNIST database of handwritten digits. Yann LeCun, Corinna Cortes, Christopher J.C. Burges. http://yann.lecun.com/exdb/mnist/

Details

This is the MNIST dataset, commonly used for benchmarking machine learning algorithms. The data is stored as a dense matrix (not sparse) since handwritten digit images have substantial non-zero content.

For NMF, it's recommended to:

  • Normalize pixel values to [0,1] by dividing by 255

  • Transpose to have pixels as rows, samples as columns

  • Use a subset for faster experimentation

The true rank is 10 (number of digit classes), though higher ranks may capture stroke variations and writing styles within digits.

Examples

# \donttest{
data(digits)
dim(digits)
#> [1] 1797   64
model <- nmf(digits, k = 10, maxit = 50)
# }