fashion mnist cloths

Usage no npm install needed!

<script type="module">
  import fashionMnist from '';


Fashion MNIST Clothes


fashion mnist clothes

The goal of this library is to provide an easy-to-use way for training and testing Fashion MNIST Clothes for neural networks (either in the browser or node.js). It includes n? different samples of fashion mnist clothes. I built this in order to work out of the box with DeepJS.


for node.js: npm install fashion-mnist --save


The most important method is fmnist.set(trainingAmount, testAmount) which takes the amount of samples for the training and test sets, and returns an object with the two sets of samples (one for training and the other one for testing). Both sets are shuffled, and there are no samples repeated in both sets.

For example:

var fmnist = require('fashion-mnist'); // this line is not needed in the browser

var set = fmnist.set(8000, 2000);

var trainingSet =;
var testSet = set.test;

That would create a random training set of 8000 fashion mnist clothes, and a test set with other random 2000 fashion mnist clothes, and there are not going to be any sample in the training set repeated in the test set.

Every set consist of an array of elements, which each of them will looks like this:

    // a 784-length array of floats representing each pixel of the 28 x 28 image, normalized between 0 and 1
    input: [0,0,0,1,1, ... ,0,0],
    // a 10-length binary array that tells which digits (from 0 to 9) is in that image
    output: [0,0,0,0,0,0,1,0,0,0]


Every digit has a set of helper functions which are listed below. To access these functions you have to target the desired digit like this: mnist[digit].

So, for example, to get a single '5' digit, you would call mnist[5].get().

These are all the helper function:


Returns a single sample of that digit, given its index. If no index is provided, a random sample is returned.

mnist[0].get(100) // [0,0,0,1,1...,0,0]


Returns the number of samples available for that digit

mnist[6].length // 1009

.range(start, end)

Returns an array of samples, corresponding to the range specified between the start and end arguments

mnist[9].range(10, 17) // [ array with 7 samples of the digit 9 ]

.set(start, end)

Returns an dataset of samples, corresponding to the range specified between the start and end arguments, every element of the array has an input and output properties, being input the normalized data for all the pixels of each image, and output a 10-length binary array representing which digit is in that image.

mnist[4].set(200, 250) // [ dataset with 50 samples of the digit 4 ready for training/testing a neural network ]


This property provides access to all the raw data for all the samples of that digit

mnist[3].raw // all the raw data for digit 3

Drawing a digit (only in the browser)

This lets you draw a given digit in a canvas context: mnist.draw(digit, context [,offsetX, offsetY])

var digit = mnist[1].get();
var context = document.getElementById('myCanvas').getContext('2d');

mnist.draw(digit, context); // draws a '1' mnist digit in the canvas

The digit provided has to be a 784-length array of normalized values (0-1). All of the following approaches will work:

var cloth = mnist[0].get(); // single cloth
var cloth = mnist[1].range(0, 100)[50]; // extracting the cloth from a range
var cloth = mnist[4].set(0, 100)[20].input; // extracting the cloth from a dataset
var cloth = mnist.set(8000, 2000).training[0].input; // extracting the cloth from a training set
var cloth = mnist.set(8000, 2000).test[15].input; // extracting the cloth from a test set


If you want to contribute feel free to submit PR's, just make sure to run npm run build to build the /dist files before submitting.