Beginning Anomaly Detection Using

Figure 3-48.  Code to generate the AUC score for this model based on the test set

Figure 3-49.  Code to see what the predictions actually look like before rounding 

them

Chapter 3   IntroduCtIon to deep LearnIng

---

109

The data values for the predictions for every other class besides the one it predicts 

correctly are so small that rounding them off is insignificant. The AUC score is shown in 

Figure 

3-51

.

That’s a really good AUC score! This score indicates that this model is really good at 

identifying handwritten digits, provided they’re in a similar format to the MNIST data set 

you used during training.

Referring back to the convolutional layers, let’s run some code to see what the feature 

maps look like after the first two convolutional layers compared to the original image.

Run the code in Figure 

3-52

 and look at the output in Figure 

3-53

.

Figure 3-50.  The output for running the code in Figure 

3-49

Figure 3-51.  The generated AUC score for the model. This is the output of running 

the code in Figure 

3-48

Chapter 3   IntroduCtIon to deep LearnIng

---

110