Beginning Anomaly Detection Using

Figure 5-64.  Code to check the five-number summary of the anomalous data

Figure 5-65.  Based on the maximum value, you don’t need to filter out any values 

for cost, except for what is an anomaly and what is a normal point

The output should look somewhat like Figure 

5-65

.

Now you can graph the free energy vs. the probabilities for each value in the test set 

separated by their label. Run the code in Figure 

5-66

.

Chapter 5   Boltzmann maChines

---

211

The output should look somewhat like Figure 

5-67

.

Figure 5-67.  There seems to be a defined separation between the anomalies and 

the normal data points. The anomalies in general seem to have a much higher free 

energy cost and a lower-than-usual probability of occurring

Figure 5-66.  Code to plot the free energy vs. the probability for each entry in the 

test set. All of the anomalies have free energies under 1500, so you can filter out all 

values for cost under 1500 to make the graph easier to visualize

Chapter 5   Boltzmann maChines

---

212

Once again, the RBM has learned the distribution well enough that there’s a clear 

and defined separation between the anomalies and the normal data entries.

 Summary

In this chapter, we discussed restricted Boltzmann machines and how they can be used 

for anomaly detection. We also explored the application of the RBM to two data sets that 

represented two cases where standardization of the data is necessary for proper training. 

You now know more about what an RBM is, how it works, and how to apply it to different 

data sets.

In the next chapter, we will take a look at anomaly detection using recurrent neural 

networks.

Chapter 5   Boltzmann maChines

---

213

© 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_6