Failed prediction: Simple square wave or similar prediction example

I am using and posted a similar question in that subforum, but after reading related questions, I came to the conclusion that it is better to post in this subforum for the type of question that I am posing as seems more related to the port of the NuPIC code.

I am trying to get a simple example working with a slightly noisy square wave where the termination of the signal is detected as an anomaly. I have played with numerous parameters for 2 days and the signal “looks” like it is being predicted except at the point where the signal stops. At that point, I would expect a prediction of the rise at the start of the square wave. As it is now, the output seems less predictive and more as if the predictor is simply following the input. This is the case both with 10 oscillations of the square wave and 100. I am wondering if there are simple examples like a square wave follower that can be used as reference?

I would be happy to add the code and documentation if I can get it working. The code being used is based on the HotGym example which uses TemporalMemory and the SpatialPooler.



1 Like