MLDatasets.jl

This package represents a community effort to provide a common interface for accessing common Machine Learning (ML) datasets. In contrast to other data-related Julia packages, the focus of MLDatasets.jl is specifically on downloading, unpacking, and accessing benchmark dataset. Functionality for the purpose of data processing or visualization is only provided to a degree that is special to some dataset.

Package Status	Build Status

This package is a part of the JuliaML ecosystem. Its functionality is build on top of the package DataDeps.jl.

Introduction

The way MLDatasets.jl is organized is that each dataset has its own dedicated sub-module. Where possible, those sub-module share a common interface for interacting with the datasets. For example you can load the training set and the test set of the MNIST database of handwritten digits using the following commands:

using MLDatasets

train_x, train_y = MNIST.traindata()
test_x,  test_y  = MNIST.testdata()

To load the data the package looks for the necessary files in various locations (see DataDeps.jl for more information on how to configure such defaults). If the data can't be found in any of those locations, then the package will trigger a download dialog to ~/.julia/datadeps/MNIST. To overwrite this on a case by case basis, it is possible to specify a data directory directly in traindata(dir = <directory>) and testdata(dir = <directory>).

Available Datasets

Check out the latest documentation

Additionally, you can make use of Julia's native docsystem. The following example shows how to get additional information on MNIST.traintensor within Julia's REPL:

?MNIST.traintensor

Each dataset has its own dedicated sub-module. As such, it makes sense to document their functionality similarly distributed. Find below a list of available datasets and links to their their documentation.

Image Classification

This package provides a variety of common benchmark datasets for the purpose of image classification.

Dataset	Classes	traintensor	trainlabels	testtensor	testlabels
MNIST	10	28x28x60000	60000	28x28x10000	10000
FashionMNIST	10	28x28x60000	60000	28x28x10000	10000
CIFAR-10	10	32x32x3x50000	50000	32x32x3x10000	10000
CIFAR-100	100 (20)	32x32x3x50000	50000 (x2)	32x32x3x10000	10000 (x2)
SVHN-2 (*)	10	32x32x3x73257	73257	32x32x3x26032	26032

(*) Note that the SVHN-2 dataset provides an additional 531131 observations aside from the training- and testset

EMNIST packages 6 different extensions of the MNIST dataset involving letters and digits and variety of test train split options. Each extension has the standard test/train data/labels nested under it as shown below.

traindata = EMNIST.Balanced.traindata()
testdata = EMNIST.Balanced.testdata()
trainlabels = EMNIST.Balanced.trainlabels()
testlabels = EMNIST.Balanced.testlabels()

Dataset	Classes	traindata	trainlabels	testdata	testlabels	balanced classes
ByClass	62	697932x28x28	697932x1	116323x28x28	116323x1	no
ByMerge	47	697932x28x28	697932x1	116323x28x28	116323x1	no
Balanced	47	112800x28x28	112800x1	18800x28x28	18800x1	yes
Letters	26	124800x28x28	124800x1	20800x28x28	208000x1	yes
Digits	10	240000x28x28	240000x1	40000x28x28	40000x1	yes
MNIST	10	60000x28x28	60000x1	10000x28x28	10000x1	yes

Misc. Datasets

Dataset	Classes	traintensor	trainlabels	testtensor	testlabels
Iris	3	4x150	150	-	-
BostongHousing	-	13x506	1x506	-	-

Language Modeling

	Train x	Train y	Test x	Test y
PTBLM	42068	42068	3761	3761
UD_English	12543	-	2077	-

PTBLM

The PTBLM dataset consists of Penn Treebank sentences for language modeling, available from tomsercu/lstm. The unknown words are replaced with <unk> so that the total vocabulary size becomes 10000.

This is the first sentence of the PTBLM dataset.

x, y = PTBLM.traindata()

x[1]
> ["no", "it", "was", "n't", "black", "monday"]
y[1]
> ["it", "was", "n't", "black", "monday", "<eos>"]

where MLDataset adds the special word: <eos> to the end of y.

Text Analysis (POS-Tagging, Parsing)

UD English

The UD_English Universal Dependencies English Web Treebank dataset is an annotated corpus of morphological features, POS-tags and syntactic trees. The dataset follows CoNLL-style format.

traindata = UD_English.traindata()
devdata = UD_English.devdata()
testdata = UD_English.devdata()

Documentation

Check out the latest documentation

Additionally, you can make use of Julia's native docsystem. The following example shows how to get additional information on MNIST.convert2image within Julia's REPL:

?MNIST.convert2image

  convert2image(array) -> Array{Gray}

  Convert the given MNIST horizontal-major tensor (or feature matrix) to a vertical-major Colorant array. The values are also color corrected according to
  the website's description, which means that the digits are black on a white background.

  julia> MNIST.convert2image(MNIST.traintensor()) # full training dataset
  28×28×60000 Array{Gray{N0f8},3}:
  [...]

  julia> MNIST.convert2image(MNIST.traintensor(1)) # first training image
  28×28 Array{Gray{N0f8},2}:
  [...]

Installation

To install MLDatasets.jl, start up Julia and type the following code snippet into the REPL. It makes use of the native Julia package manger.

import Pkg
Pkg.add("MLDatasets")

License

This code is free to use under the terms of the MIT license.