Beginning Anomaly Detection Using

val_loss: 0.0248 - val_mean_absolute_error: 0.0248

Now you can use the predicted dataset and the test dataset to compute the difference 

as diff, which is then passed through vector norms. Calculating the length or magnitude 

of vectors is often required directly as a regularization method in machine learning. 

Then you can sort the scores/diffs and use a cutoff value to pick the threshold. This 

obviously can change as per the parameters you choose, particularly the cutoff value 

(which is 0.99 in Figure 

6-33

). The figure also shows the code to compute the threshold.

Figure 6-32.  Code to predict on the testing dataset

Chapter 6   Long Short-term memory modeLS 

---

240

You got 0.333 as the threshold; anything above is considered an anomaly.

Figure 

6-34

 shows the code to plot testing dataset (GREEN) and the corresponding 

predicted dataset (RED).