mnists/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, using default mnists.prepare parameters:

dataset	train images shape	train labels shape	test images shape	test labels shape
emnist_balanced	(112800, 1, 28, 28)	(112800, 47)	(18800, 1, 28, 28)	(18800, 47)
emnist_byclass	(697932, 1, 28, 28)	(697932, 62)	(116323, 1, 28, 28)	(116323, 62)
emnist_bymerge	(697932, 1, 28, 28)	(697932, 47)	(116323, 1, 28, 28)	(116323, 47)
emnist_digits	(240000, 1, 28, 28)	(240000, 10)	(40000, 1, 28, 28)	(40000, 10)
emnist_letters	(124800, 1, 28, 28)	(124800, 26)	(20800, 1, 28, 28)	(20800, 26)
emnist_mnist	(60000, 1, 28, 28)	(60000, 10)	(10000, 1, 28, 28)	(10000, 10)
fashion_mnist	(60000, 1, 28, 28)	(60000, 10)	(10000, 1, 28, 28)	(10000, 10)
mnist	(60000, 1, 28, 28)	(60000, 10)	(10000, 1, 28, 28)	(10000, 10)