Beginning Anomaly Detection Using

IRU

HSRFK

LQ UDQJH

QXPBHSRFKV 

WUDLQ PRGHOGHYLFHWUDLQBORDGHUFULWHULRQRSWLPL]HUHSRFK

7HVWLQJSKDVH

WHVW PRGHOGHYLFHWHVWBORDGHU

Figure B-12.  Initializing the model and passing it to the GPU, defining your 

criterion function (cross entropy loss), and defining your optimizer (the Adam 

optimizer). Then, the training and testing functions are called

appendix B   intro to pytorch

---

374

Figure B-13.  What the code from Figures 

B-11

 and 

B-12

 should look like after 

pasting them into a Jupyter cell

appendix B   intro to pytorch

---

375

After the training process, you get Figure 

B-14

 and Figure 

B-15

.

Figure B-14.  The initial output of the training process

Figure B-15.  The training process has finished

Although in your Keras examples you didn’t spread apart your training and testing 

functions (since they’re just one line each), more complicated implementations of 

models involving custom layers, models, and so on can be formatted in a similar fashion 

to the PyTorch example above.

appendix B   intro to pytorch

---

376

Hopefully, you understand a bit more on how to implement, train, and test neural 

networks in PyTorch.

Next, we will explain some of the basic functionality that PyTorch offers in terms 

of model layers (activations included), loss functions, and optimizers, and then you’ll 

explore PyTorch applications of temporal convolutional neural networks to the data set 

found in Chapter 

7

.