In this chapter, we will explore various datasets available by default in Keras and how to load and use them.

Load the CIFAR-10 small images classification dataset from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz. The CIFAR-10 dataset is made up of 60,000 32 x 32 color images in 10 classes, and there are 6000 images per class. The dataset consists of 50,000 training images and 10,000 test images.

The dataset has been divided into five training batches and one test batch, each with 10,000 images. The test batch contains 1,000 randomly selected images from each class. The training batches contain the rest of the images in a random order; some training batches may contain more images from one class than another. The training batches contain 5,000 images from each class, such as shown in the following image:

Reference: https://www.cs.toronto.edu/~kriz/cifar.html.

A training dataset of 50,000 32 x 32 pixel color images labeled over 100 categories and 10,000 test images, this dataset is similar to CIFAR-10, but it has 100 classes with 600 images in each class. Five-hundred training images and 100 testing images are in each class. The 100 classes in CIFAR-100 are grouped into 20 superclasses. Each image comes with a coarse label (the superclass to which it belongs) and a fine label (the class to which it belongs).
A list of classes in CIFAR-100 is as follows:

MNIST is a dataset of 60,000 28 x 28 pixel grayscale images of 10 digits. It also contains a test set of 10,000 images. The dataset consists of the following four files:

• train-images-idx3-ubyte.gz: Training set images (9912422 bytes), go to http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz
• train-labels-idx1-ubyte.gz: Training set labels (28881 bytes), go to http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz
• t10k-images-idx3-ubyte.gz: Test set images (1648877 bytes), go to http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz
• t10k-labels-idx1-ubyte.gz: Test set labels (4542 bytes), go to http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz

Data in these files is stored in the IDX format. The IDX file format is a format for vectors and multidimensional matrices of various numerical types. You can find more info...

Keras can take data directly from a numpy array in addition to preexisting datasets.

Let's take an existing .csv file from the internet and use it to create a Keras dataset:

dataset = numpy.loadtxt("https://raw.githubusercontent.com/jbrownlee/Datasets/master/pima-indians-diabetes.data.csv", delimiter=",")
# split into input (X) and output (Y) variables
X = dataset[:,0:8]
Y = dataset[:,8]

Note that the dataset can be directly loaded from the URL with the .csv file.

The output of the preceding code is listed in the following snippet:

[ 6. 148. 72. 35. 0. 33.6 0.627 50. ]
1.0

Let's look at creating a basic model in Keras.

Model is a subclass of Network. The Model class adds training and evaluation routines to a Network. The following diagram shows the relationship between the various classes.

There are two types of models in Keras:

• Sequential models
• Models created using functional APIs

A Sequential model can be created by passing a stack of layers to the constructor of a class called Sequential. 

Creating a basic Sequential mode involves specifying one or more layers.

We will create a Sequential network with four layers.

1. Layer 1 is a dense layer which has input_shape of (*, 784) and an output_shape of (*, 32)

Multiple layers in Keras can share the output from one layer. There can be multiple different feature extraction layers from an input, or multiple layers can be used to predict the output from a feature extraction layer.

Let's look at both of these examples.

In this section, we show how multiple convolutional layers with differently sized kernels interpret an image input. The model takes colored CIFAR images with a size of 32 x 32 x 3 pixels. There are two CNN feature extraction submodels that share this input; the first has a kernel size of 4, the second a kernel size of 8. The outputs from these feature extraction sub-models are flattened into vectors and...