CIFAR 10 Library for Javascript

As you know, CIFAR-10 is famous library in Deep Learning. In python, this library is easy to test, while it is difficult to manuplate it in Javascript.

Features

• based on stream : use less memory
• exclude data file : only adapter without data files
• general purpose : this is not design for only CIFAR-10

How to use

This is quite simple, but you should run the get_datasets.sh, which downloads an actual data file(CIFAR-10 binary version).

$ ./node_modules/cifar-10/get_datasets.sh

It will download data files and this script use wget. After that, you can access data by load().

const CIFAR10 = require('cifar-10');
const { X_train, y_train, X_test, y_test } = await CIFAR10.load();

// or callback style

CIFAR10.load().then(function(data) {
  // data = { X_train, y_train, X_test, y_test }

});

APIs

Load(option)

Return new promise to process data from given option.

option = {
  totalSize,    // the length of bytes for each case
  mapper,       // binary to matrix mapper
  trainFiles,   // array for files to be a training set
  testFiles,    // array for files to be a testing set
}

For example, this example is for CIFAR 10.

Contributing

PR or issue reporting are always welcome. Feel free to report or advice to this project.