Hi Yajingfu, I think your understanding of how it works seems pretty good. Please see this sample code on how to use it:
https://github.com/subutai/nupic.subutai/blob/master/run_anomaly/run_anomaly.py That code is almost identical to how we use it in Grok. --Subutai On Thu, Feb 5, 2015 at 5:32 AM, 天朗气清 <[email protected]> wrote: > hi > I'm studying anomaly detection of NuPIC. But I don't know how to > calculated anomaly likelihood clearly. From some vedio and webpages, It > seems like that anomaly likelihood is a supplement of anomalyscore. > Traditional offline anomaly detection method can calculate anomaly from > back to front. But online anomaly detection method can't do that. So > anomaly likelihood is a reference of anomaly score, to detect whether the > high score is really anomaly or not. I don't know whether my opinion is > right or not. > I read the sourcecode about anomaly likelihood but I'm sorry that I can't > understand the specific steps about computing it. How often to calculate > it? We get a normal distribution of anomaly score, and likelihood is a > probability that current anomaly score bigger than movingaverage. Is it > right? Can you tell me in detail? > Hoping for your reply. > > --yajingfu >
