A quick note to those interested in anomaly detection. I gave a talk at the last hackathon on using CLA’s for anomaly detection. A link to the talk is below [1].
During the talk I discussed how (and why) we convert raw CLA anomaly scores into a “likelihood” score. This is the score that is actually used by Grok the product. The scheme seems to work well on the datasets that we have tried, particularly the really noisy ones. We decided to release that code to the community since it might be useful to others. For now I have put the code in my repository [2]. I have updated the run_anomaly script in there to use the likelihood code. Hopefully it should be easy to use now. If people want, we can put it into NuPIC (as an optional routine) in which case I’ll create a PR for merging it in. —Subutai [1] http://numenta.org/blog/#sessions [2] https://github.com/subutai/nupic.subutai/tree/master/run_anomaly
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
