Red Datasets

Description

Red Datasets provides classes that provide common datasets such as iris dataset.

You can use datasets easily because you can access each dataset with multiple ways such as #each and Apache Arrow Record Batch.

Install

% gem install red-datasets

Available datasets

• Adult Dataset
• Aozora Bunko
• California Housing
• CIFAR-10 Dataset
• CIFAR-100 Dataset
• CLDR language plural rules
• Communities and crime
• Diamonds Dataset
• E-Stat Japan
• Fashion-MNIST
• Fuel Economy Dataset
• Geolonia Japanese Addresses
• Hepatitis
• House of Councillors of Japan
• House of Representatives of Japan
• Iris Dataset
• Libsvm
• MNIST database
• Mushroom
• Penguins
• The Penn Treebank Project
• PMJT - Pre-Modern Japanese Text dataset list
• Postal Codes in Japan
• Rdatasets
• Seaborn
• Sudachi Synonym Dictionary
• Wikipedia
• Wine Dataset

Usage

Here is an example to access Iris Data Set by #each or Table#to_h or Table#fetch_values.

require "datasets"

iris = Datasets::Iris.new
iris.each do |record|
  p [
     record.sepal_length,
     record.sepal_width,
     record.petal_length,
     record.petal_width,
     record.label,
  ]
end
# => [5.1, 3.5, 1.4, 0.2, "Iris-setosa"]
# => [4.9, 3.0, 1.4, 0.2, "Iris-setosa"]
  :
# => [7.0, 3.2, 4.7, 1.4, "Iris-versicolor"]


iris_hash = iris.to_table.to_h
p iris_hash[:sepal_length]
# => [5.1, 4.9, .. , 7.0, ..
p iris_hash[:sepal_width]
# => [3.5, 3.0, .. , 3.2, ..
p iris_hash[:petal_length]
# => [1.4, 1.4, .. , 4.7, ..
p iris_hash[:petal_width]
# => [0.2, 0.2, .. , 1.4, ..
p iris_hash[:label]
# => ["Iris-setosa", "Iris-setosa", .. , "Iris-versicolor", ..


iris_table = iris.to_table
p iris_table.fetch_values(:sepal_length, :sepal_width, :petal_length, :petal_width).transpose
# => [[5.1, 3.5, 1.4, 0.2],
      [4.9, 3.0, 1.4, 0.2],
      :
      [7.0, 3.2, 4.7, 1.4],
      :

p iris_table[:label]
# => ["Iris-setosa", "Iris-setosa", .. , "Iris-versicolor", ..

Here is an example to access The CIFAR-10/100 dataset by #each:

CIFAR-10

require "datasets"

cifar = Datasets::CIFAR.new(n_classes: 10, type: :train)
cifar.metadata
#=> #<struct Datasets::Metadata name="CIFAR-10", url="https://www.cs.toronto.edu/~kriz/cifar.html", licenses=nil, description="CIFAR-10 is 32x32 image dataset">licenses=nil, description="CIFAR-10 is 32x32 image datasets">
cifar.each do |record|
  p record.pixels
  # => [59, 43, 50, 68, 98, 119, 139, 145, 149, 143, .....]
  p record.label
  # => 6
end

CIFAR-100

require "datasets"

cifar = Datasets::CIFAR.new(n_classes: 100, type: :test)
cifar.metadata
#=> #<struct Datasets::Metadata name="CIFAR-100", url="https://www.cs.toronto.edu/~kriz/cifar.html", licenses=nil, description="CIFAR-100 is 32x32 image dataset">
cifar.each do |record|
  p record.pixels
  #=> [199, 196, 195, 195, 196, 197, 198, 198, 199, .....]
  p record.coarse_label
  #=> 10
  p record.fine_label
  #=> 49
end

MNIST

require "datasets"

mnist = Datasets::MNIST.new(type: :train)
mnist.metadata
#=> #<struct Datasets::Metadata name="MNIST-train", url="http://yann.lecun.com/exdb/mnist/", licenses=nil, description="a training set of 60,000 examples">

mnist.each do |record|
  p record.pixels
  # => [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, .....]
  p record.label
  # => 5
end

NArray compatibility

• red-datasets-numo-narray

How to develop Red Datasets

1. Fork https://github.com/red-data-tools/red-datasets
2. Create a feature branch from master
3. Develop in the feature branch
4. Pull request from the feature branch to https://github.com/red-data-tools/red-datasets

License

The MIT license. See LICENSE.txt for details.