Beginning Anomaly Detection Using

Figure 4-43.  Code to import packages

Figure 4-44.  Code to load MNIST images

Chapter 4   autoenCoders

---

155

Split the dataset into training and testing subsets. Also, reshape the data to 28X28 

images. Figure 

4-45

 shows the code to load and reshape images.

Figure 

4-46

 shows the code to display the images.

Figure 4-45.  Code to load and reshape images

Figure 4-46.  Code to display the images

Create a CNN model with Convolutions and MaxPool layers. Figure 

4-47

 shows the 

code to create the neural network.

Chapter 4   autoenCoders

---

156

Compile the model using RMSprop as the optimizer and mean squared error for the 

loss computation. The RMSprop optimizer is similar to the gradient descent algorithm 

with momentum. Figure 

4-48

 shows the code to compile the model.

Figure 4-47.  Code to create the neural network

Chapter 4   autoenCoders

---

157

Now, you can start training the model using the training dataset to validate the model 

at every step. Choose 32 as the batchsize and 20 epochs. The training process outputs 

the loss and accuracy as well as the validation loss and validation accuracy at each epoch. 

Figure 

4-49

 shows the code to start training the model.