Beginning Anomaly Detection Using

IURP

VNOHDUQPHWULFV

LPSRUW

URFBDXFBVFRUH

SUHGV 7&1SUHGLFW [BWHVW

DXF URFBDXFBVFRUH QSURXQG SUHGV \BWHVW

SULQW

$8&^`IRUPDW DXF

Figure 7-64.  Code to check the AUC score given the rounded predictions and the 

test sets

Chapter 7   temporal Convolutional networks

---

295

That’s a nice AUC score! So for both the encoder-decoder TCN and dilated TCN 

architectures, you’ve managed to attain AUC scores of over 98% on the credit card 

data set in a supervised setting. Although both models trained and performed in a 

supervised setting, since the anomalies and the normal entries were labeled as such, 

the key takeaway is that TCNs are incredibly quick to train with GPUs and can perform 

really well.

 Summary

In this chapter, we discussed temporal convolutional networks and showed how they 

fare when applied to anomaly detection.

In the next chapter, we will look at practical use case of anomaly detection.

Figure 7-65.  The generated AUC score

Chapter 7   temporal Convolutional networks

---

297

© 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_8