README.md

In jlmelville/mnist: Code to Download the MNIST Database of Handwritten Digits

mnist

NOTE: This package has been superseded by snedata. See you over there.

mnist is an R package to download the MNIST database, based on a gist by Brendan O'Connor.

The entire dataset is returned as a single data frame. The first 60,000 instances are the training set, the remaining 10,000 the test set. The pixel values (integers in the range 0-255) are in columns with name px1, px2, px3 etc. The label representing the numerical value of the digit is in the Label column (which is stored as a factor).

Installing:

# install.packages("devtools")
devtools::install_github("jlmelville/mnist")
library(mnist)

Using:

# fetch the data set from the MNIST website
mnist <- download_mnist()

# view the fifth digit
show_digit(mnist, 5)

# first 60,000 instances are the training set
mnist_train <- head(mnist, 60000)
# the remaining 10,000 are the test set
mnist_test <- tail(mnist, 10000)

# PCA on 1000 random training examples
mnist_r1000 <- mnist_train[sample(nrow(mnist_train), 1000), ]

pca <- prcomp(mnist_r1000[, 1:784], retx = TRUE, .rank = 2)
# plot the scores of the first two components
plot(pca$x[, 1:2], type = 'n')
text(pca$x[, 1:2], labels = mnist_r1000$Label, cex = 0.5,
  col = rainbow(length(levels(mnist_r1000$Label)))[mnist_r1000$Label])

# save data set to disk
save(mnist, file = "mnist.Rda")

License

This package is licensed under the MIT License.

See Also

• A similar project by Xavier Robin.
• I have similar R packages for simulation, Olivetti and Frey Faces and COIL-20 datasets.
• For doing an embedding, you could give sneer a go.

jlmelville/mnist documentation built on May 19, 2019, 12:47 p.m.