Beginning Anomaly Detection Using

Figure 3-42.  A graph showing the ReLU function

Figure 3-43.  Formula for the softmax activation function

Chapter 3   IntroduCtIon to deep LearnIng

---

104

Figure 3-44.  The output for the code in Figure 

3-33

. Note how it tells you the 

output shapes of each layer and the number of parameters; this can be useful when 

creating custom models and finding out that there is a mismatch between the 

dimensionality of what a layer expects and what it actually receives

Chapter 3   IntroduCtIon to deep LearnIng

---

105

Now let’s move on to training the data. Depending on your setup, this can take 

anywhere from a few seconds to several minutes. Without cuda, expect that this will take 

much longer.

Run the code in Figure 

3-45

.

FKHFNSRLQW 0RGHO&KHFNSRLQW ILOHSDWK NHUDVB01,67B&11K

YHUERVH 

VDYHBEHVWBRQO\ 7UXH

PRGHOILW [BWUDLQ\BWUDLQ

EDWFKBVL]H EDWFKBVL]H

HSRFKV QBHSRFKV

YHUERVH 

YDOLGDWLRQBGDWD [BWHVW\BWHVW 

FDOOEDFNV >FKHFNSRLQW@

VFRUH PRGHOHYDOXDWH [BWHVW\BWHVWYHUERVH 

SULQW 7HVWORVV VFRUH>@

SULQW 7HVWDFFXUDF\ VFRUH>@

Figure 3-45.  Code to train the model and print accuracy and loss values for the 

test set

Chapter 3   IntroduCtIon to deep LearnIng

---

106

PRGHOILW [BWUDLQ\BWUDLQ

EDWFKBVL]H EDWFKBVL]H

HSRFKV QBHSRFKV

YHUERVH 

YDOLGDWLRQBGDWD [BWHVW\BWHVW

VFRUH PRGHOHYDOXDWH [BWHVW\BWHVWYHUERVH 

SULQW 7HVWORVV VFRUH>@

SULQW 7HVWDFFXUDF\ VFRUH>@

Figure 3-46.  Run this code if you don’t want to save the model

The variable checkpoint will store the model in the same folder as this code with 

the name keras_MNIST_CNN.h5. If you don’t want to save the model, run the code in 

Figure 

3-46

 instead.

Chapter 3   IntroduCtIon to deep LearnIng

---

107

Figure 3-47.  The output of running the training function, accompanied by the loss 

and accuracy values for the test set

If successful, you should see something like Figure 

3-47

.

Let’s check the AUC score for this. Run the code in Figure 

3-48

.

Chapter 3   IntroduCtIon to deep LearnIng

---

108

Basically, the variable predictions are a list of arrays with 10 elements, each containing 

the probability values for class predictions for each of the x_test data samples.

To check the values for the predictions before doing np.round(), run the code in 

Figure 

3-49

 and see the results in Figure 

3-50

.

IURPVNOHDUQPHWULFVLPSRUWURFBDXFBVFRUH

SUHGV PRGHOSUHGLFW [BWHVW

DXF URFBDXFBVFRUH QSURXQG SUHGV \BWHVW

SULQW $8&^`IRUPDW DXF