Help debugging for anomaly detection


My approach was the following:

  1. Vet any differences between all the individual algorithms.
  2. Vet any differences between raw-chained components and the Network API to rule out NAPI perturbation.
  3. Vet any differences in actual realized configuration parameters.

All the above checked out. I reviewed my approach with my manager Gerald at, and he believes as well as I that the next step is to try swarming over the params?


Also Matt, I’m not sure how your time is - but if you have time, I’d like to update the examples repo at some point (today or soon) so you can try your visualization on the most recent code? Again, when you have chance…



For all those who have been following the issue of HTM.Java receiving its final vetting, it is now fully vetted and will be fully operational following the next upcoming release (0.6.9).

Thanks to @mrcslws (Marcus Lewis) identifying a parameter that gets set internally within the OPF and NuPIC Network API ( potentialRadius), which for global inhibition (the most frequently used mode of operation for the SpatialPooler); is set to the inputWidth internally instead of its default of “16” - we now have respectable scores within NAB, and Anomaly Detection will no longer be broken. See Marcus’ Blog about the significance of this parameter!

YAY, Marcus!!!

HTM.Java has undergone 2 full re-writes of its TemporalMemory and test, 1 re-write of its SpatialPooler and test, the addition of multiple compatibility and consistency tests - and one full toss out the window! :stuck_out_tongue: And is now ship-shape!

Thanks also go out to @fergalbyrne (Fergal Byrne), who heroically stepped right up to help smooth over rough spots within the code; @alavin (Alex Lavin), who’s dedication, perseverance and guidance also contributed greatly; @lscheinkman (Luiz Sheinkman) who’s enthusiasm caused him to write the Java Anomaly Detector in the first place; (Matt Taylor) without whom, nothing would get done anywhere :-P; (Subutai Ahmad) who’s early direction contributed greatly; (Gerald Loeffler - Head of Development at, who’s advice and never-wavering support kept me going; and (Scott Purdy) who’s advice and general vast knowledge of all things NuPIC, was relied upon very frequently (in the background as his modesty dictates).

Thank you to all the interested parties who have been keeping up with this issue and silently cheering by following closely as indicated by the popularity of this forum post. Your interest was also inspiring for all of us to persevere!


P.S. I only linked 4 users because apparently the Forum is set up to limit user-mentioning to 4 within a post?



Delighted to hear this, @cogmission. People can just scroll up here and also on Github to see how many people have chipped in on this effort (I only spent a couple of days on it). While one of the motivations for many of us was to make work well, perhaps this whole saga just shows the power of so many people in this community helping build the barn using their own particular talents to the best. Biggest congrats must go to both yourself and @rhyolight for everything you do to energise and befriend a very diverse community.


Yes, many people chipped in, and it took a long time, but don’t we all know each other so much better now? :wink: