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	Package Evaluator	Build Status

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

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>).

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.

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

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

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.

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()

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}:
  [...]

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")

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