Beginning Anomaly Detection Using

Figure 4-66.  Code to show the neural network

Figure 4-67.  Code to compile the model

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

loss computation. Adam is an optimization algorithm that can be used instead of the 

classical stochastic gradient descent procedure to update network weights iteratively 

based on training data. Figure 

4-67

 shows the code to compile the model.

Chapter 4   autoenCoders

---

172

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 

4-68

 shows the code to train the model.

Figure 4-68.  Code to train the model

Chapter 4   autoenCoders

---

173

Now that the training process is complete, let’s evaluate the model for loss and 

accuracy. Figure 

4-69

 shows that the accuracy is 0.23. It also shows the code to evaluate 

the model.

The next step is to calculate the errors, and detect and also plot the anomalies and 

the errors. Choose a threshold of 10. Figure 

4-70

 shows the code to predict the anomalies 

based on the threshold.