HTM anomaly raw score threshold on multi features problems

I create HTM model with multi features input for anomaly detection problem.
In HTM framework we can’t predict multi features output, only calculate the raw score.
May someone have a suggestion how to calculate the score threshold?

You want to run an AnomalyLikelihood calculation, then apply a very high threshold like 0.9999. Watch this: Live Q/A session on Anomaly Detection.