downloads and prepares various mnist-compatible datasets

Go to file

Connor Olding 25073601b7 make bold text into headers and note anaconda		2018-03-15 01:32:21 +01:00
mnists	split hashes to separate file	2018-03-15 01:27:01 +01:00
.gitignore	add .gitignore	2018-03-15 01:26:10 +01:00
LICENSE	init	2018-03-14 16:45:53 +01:00
README.md	make bold text into headers and note anaconda	2018-03-15 01:32:21 +01:00
requirements.txt	init	2018-03-14 16:45:53 +01:00
setup.py	init	2018-03-14 16:45:53 +01:00
TODO	init	2018-03-14 16:45:53 +01:00

README.md

mnists

downloads and prepares various mnist-compatible datasets.

files are downloaded to ~/.mnist and checked for integrity by SHA-256 hashes.

dependencies

python 3.6 (or later), numpy.

install

pip install --upgrade --upgrade-strategy only-if-needed 'https://github.com/notwa/mnists/tarball/master#egg=mnists'

I've added --upgrade-strategy to the command-line so you don't accidentally "upgrade" numpy to a version not compiled specifically for your system. This can happen when using e.g. Anaconda.

usage

import mnists

dataset = "emnist_balanced"
train_images, train_labels, test_images, test_labels = mnists.prepare(dataset)

the default images shape is (n, 1, 28, 28). pass flatten=True to mnists.prepare to get (n, 784).

datasets

in alphabetical order:

emnist

emnist_balanced
emnist_byclass
emnist_bymerge
emnist_digits
emnist_letters
emnist_mnist

fashion-mnist

fashion_mnist

mnist

mnist