Beginning Anomaly Detection Using

Figure 3-61.  The same process repeats, leading to a new value for x

Figure 3-62.  Once again, you redefine x as x

″. The process then continues for as 

many layers as you have in the network

Chapter 3   IntroduCtIon to deep LearnIng

---

118

This is the same logic behind the rest of the code, where the output of the activation 

layer for the old is now the new definition of x. This new x then goes to the next layer, 

where a function is applied after it goes through a layer and then that data becomes the 

new definition of x, and so on.

So

x = x.view(-1, 12*12*64)

performs the same function as the flatten layer in the Keras example.

Now you can move on to training your data (Figure 

3-63

).

PRGHO &11 WR GHYLFH

FULWHULRQ QQ&URVV(QWURS\/RVV

RSWLPL]HU WRUFKRSWLP$GDP PRGHOSDUDPHWHUV OU OHDUQLQJBUDWH

WRWDOBVWHS OHQ WUDLQBORDGHU