Beginning Anomaly Detection Using

Figure 5-27.  The output of training the model

Chapter 5   Boltzmann maChines

---

193

Figure 5-28.  Code to get the costs from the test set and get the AUC scores from that

The output should be something like Figure 

5-29

.

Figure 5-29.  The AUC score ended up at 95.84%

Considering the seemingly simple architecture of the RBM (with how few nodes 

there are in the model compared to neural networks), that’s a pretty good AUC score!

You can also graph the free energy vs. the probability of each data point to get an 

idea of what the anomalies look like compared to the normal data points. Before you do 

that, let’s check a five-number summary of each data set to get a sense of how they are 

distributed.

Figure 

5-30

 shows the code for the five-number summary of the normal data.