Beginning Anomaly Detection Using

Figure 7-35.  Defines all of the one-dimensional convolutional layers and the 

dropout layers in the model

Chapter 7   temporal Convolutional networks

---

278

The code chunk in Figure 

7-36

 defines the last two layers, which consist of a layer to 

flatten the data and one layer to represent the two classes.

Now let’s compile the model and look at the summary of the layers (see Figure 

7-37

).

The output should look like Figure 

7-38

.

)ODWWHQOD\HUWRIHHGLQWRWKHRXWSXWOD\HU

IODW )ODWWHQ GURSB

RXWSXWBOD\HU 'HQVH DFWLYDWLRQ VRIWPD[ IODW

7&1 0RGHO LQSXWV LQSXWBOD\HURXWSXWV RXWSXWBOD\HU

Figure 7-36.  Defines the last two layers, which consist of a layer to flatten the data 

and one layer to represent the two classes

7&1FRPSLOH ORVV NHUDVORVVHVFDWHJRULFDOBFURVVHQWURS\

RSWLPL]HU RSWLPL]HUV$GDP OU  

PHWULFV >µPDH¶  DFFXUDF\ @

FKHFNSRLQWHU 0RGHO&KHFNSRLQW ILOHSDWK PRGHOB7&1BFUHGLWFDUGK

YHUERVH 

VDYHBEHVWBRQO\ 7UXH

7&1VXPPDU\

Figure 7-37.  Code to compile the data, define a callback to save the model under 

the given filepath, and output the summary of the model

Chapter 7   temporal Convolutional networks

---

279

Looking at the model summary can help you understand more about what’s 

going on at each layer. Sometimes, it can help with debugging, where there can be 

dimensionality reductions that you don’t expect. For example, sometimes when odd 

dimensions become reduced by a factor of 2, they might become rounded down. When 

expanding back up, this can prove to be problematic because the new dimension does 

not match the old dimension. You can expect to run into problems like these with 

autoencoders, where the entire aim of the architecture is to compress the data and 

attempt to reconstruct it.

Figure 7-38.  The summary of the model. You can use this to help debug your 

models when you’re creating one from scratch by checking that the output shapes 

for the layers match the input shapes of the subsequent layer

Chapter 7   temporal Convolutional networks

---

280

Run the code in Figure 

7-39

 to begin the training process.

You should see something like Figure 

7-40

 during the training process.

At the end, you should see something like Figure 

7-41

.

Now that the training is finished, you can evaluate your model’s performance  

(see Figure 

7-42

).

TCN.fit(x_train, y_train,

batch_size=128,

epochs=25,

verbose=1,

validation_data=(x_test, y_test),

callbacks = [checkpointer])