HTM Parameters for Dummies


I might have missed it, so apologies if this is covered elsewhere. Is it possible to get a summary sheet describing the parameters by their effects on predictions or anomaly detection?

I’ve watched the HTM School videos and done some of the reading. The closest I saw was the boosting video showing the difference in performance and why. The SDR sizing video was great too.

I’m a little less sure of the numbers to use for permanence increment and decrement, max new synapse count, and duty cycles. I get the gist of what they are doing, but it’s not 100% clear how they impact the resulting network runs and what are the right ranges of numbers to use.

It’d be great to have an HTM Parameters for Dummies guide to start from like the transform guide attached.


I’ve been working on documentation a lot lately. I know this is an area we need improvement. This could change in the future, but this page should give an explanation of most model parameters, even if they are just comments alongside the values for the time being.


At the ends of the BAML (Biological and Machine Intelligence) pseudocode docs there are some good descriptions of what the parameters do. Here’s the one for Temporal Memory:

And here’s the one for Spatial Pooling:

Parameter Optimization and identification

These are great resources, thanks. They certainly help neck down where to focus. Thanks @sheiser1.


Certainly @jon, glad to help :slight_smile: