Multiclass classification with the CIFAR-10 dataset

The CIFAR-10 dataset consists of 60,000 32x32 colorful images in 10 classes, with 6,000 images per class. There are 50,000 training images and 10,000 test images.

Getting and previewing the dataset

Similar to the preceding example, we will use the MLDatasets package to retrieve the CIFAR10 dataset. Let's start by loading the package and having a quick look at the data:

using Images, ImageView, MLDatasets, MXNet

train_x, train_y = CIFAR10.traindata()
test_x, test_y = CIFAR10.testdata()
size(train_x)
# Main> (32, 32, 3, 50000)

size(train_y)
# Main> (50000,)

join(unique(train_y), ", ")
# Main> "6, 9, 4, 1, 2, 7, 8, 3, 5, 0"

We have used the size function to show the dimensionality of the data and the unique function to list the possible classes. Running the size function on the train_x dataset tells us that we have a training dataset of 500,000 32x32-pixel, 3-channel RGB images. train_y also contains 50,000 images and specifies 10...