Implement real-time anomaly detection


#1

Hi, I am new to HTM and Nupic. I was researching on implementing real-time anomaly detection on my own data previously and I came across the paper named “Unsupervised real-time anomaly detection for streaming data” published by Subutai. After briefly going through the documents, I get a little bit confused and haven’t find a solid example using real-time anomaly detection. Which API should I refer to? There are OPF API, ‘Network API’, and Algorithm API, I only found some info on this link.

However, is there any fully-described implemented code to do real-time anomaly detection? Thanks a lot.


#2

An old tutorial, but probably still helpful:

The API has been updated a bit, but the corresponding code examples in the NuPIC codebase still work.


#3

Thanks for your instant reply! Additionally, I wonder if you have existing code example for anomaly detection in real-time as the paper described?


#4

Yes! See the link I gave you above. It is an example of real-time anomaly detection.


#5

I just re-watched it for the first time in awhile, and realized how much better I understand the code now :grin:

I would also submit a request for another just like it, but where the data is being pulled from a live source rather than a stored csv file? I know NuPIC handles the data online either way, so just curious to see it being hooked into the outside world. This has been done of course, it’d just be super cool if I could see you laying the pipes for it on video (if I’m not along in this request of course).

Thanks again!