Learning the architecture of a CNN classifier

The CNN classifier covered in this chapter has two convolution layers followed by two fully connected layers in the end, in which the last layer acts as a classifier using the softmax activation function.

Getting ready

The recipe requires the CIFAR-10 dataset. Thus, the CIFAR-10 dataset should be downloaded and loaded into the R environment. Also, images are of size 32 x 32 pixels.

How to do it...

Let's define the configuration of the CNN classifier as follows:

1. Each input image (CIFAR-10) is of size 32 x 32 pixels and can be labeled one among 10 classes:

# CIFAR images are 32 x 32 pixels.
img_width  = 32L
img_height = 32L

# Tuple with height and width of images used to reshape arrays.
img_shape = c(img_width, img_height)
# Number of classes, one class for each of 10 images
num_classes = 10L

2. The images of the CIFAR-10 dataset have three channels (red, green, and blue):

# Number of color channels for the images: 3 channel for red, blue, green scales....