Beginning Anomaly Detection Using

IRU

HSRFK

LQ UDQJH

QXPBHSRFKV 

WUDLQ PRGHOGHYLFH[BWUDLQ\BWUDLQFULWHULRQ

RSWLPL]HUHSRFK

Figure B-51.  Initializing the TCN model, defining the criterion as the cross 

entropy loss, and defining the optimizer (Adam optimizer)

appendix B   intro to pytorch

---

407

The output should look somewhat like Figure 

B-52

.

7HVWLQJSKDVH

WHVW PRGHOGHYLFH[BWHVW

\BWHVW

Figure B-53.  Calling the test function

Figure B-52.  The output of the training process

And now you can evaluate your model (see Figure 

B-53

).

appendix B   intro to pytorch

---

408

The output should look somewhat like Figure 

B-54

.

Figure B-54.  The output AUC value of the testing function

With the end of this example, you will have created a TCN in both Keras and PyTorch. 

This way, you’ll have a good way to compare how the model is built, trained, and 

evaluated in both frameworks, allowing you to observe the similarities and differences in 

how both frameworks handle those processes.

By now, you should have a better understanding of how PyTorch works, especially 

with how it’s meant to be more intuitive. Think back to the training function and the 

process of converting the data sets, passing them through the GPU and through the 

model, calculated the gradients, and backpropagating. Though it’s not abstracted away 

from you like in Keras, it still makes logical sense in that the functions called directly 

correlate to the training process of a neural network.

 Summary

PyTorch is a low-level tool that allows you to quickly create, train, and test your 

own deep learning models, although it is more complicated than doing the same in 

Keras. However, it offers you much more functionality, flexibility, and customizability 

compared to Keras, and compared to TensorFlow, it is much lighter on syntax. With 

PyTorch, you don’t have to worry about switching frameworks as you get more advanced 

because of the functionality that it offers, making it a great tool to use when conduct 

deep learning research. PyTorch should be enough for most of your needs as you 

become more experienced with deep learning, and using either PyTorch or TensorFlow 

(or tf.keras + TensorFlow) is just a matter of personal preference.

appendix B   intro to pytorch

---

409

© Sridhar Alla, Suman Kalyan Adari 2019 

S. Alla and S. K. Adari, Beginning Anomaly Detection Using Python-Based Deep Learning,  

https://doi.org/10.1007/978-1-4842-5177-5