Anomality score / likehood after model load

anomaly-detection
opf
serialization

#1

Hello,

One question to the community about bit strange behaviour for my anomality detector. I use cron to kick model run on new batch of data instead of continious run. I noticed that likehood goes to 0.5 after first model load (OPF used). Is there any restriction or special parameterization for HTM model save/load in part of anomality detecting?


#2

The AnomalyLikelihood is not serialized as part of a NuPIC model when you save it. However, you can pickle it when you save the model. When you load the model again, load the anomaly likelihood and continue using it.


#3

Thanks a lot for the fast reply. I just noticed that AnomalyLikehood is an own variable and realized it was kind of stupid question. Anyways, it might help other newbies.


#4

No, not at all! It is not evident.