David Keeney

htm.core is a rework of NuPic so that it will work. I think you will find that nearly everything you want to do with NuPic should be covered by htm.core....

The rest_server is a separate executable program. It is something that python can make a LAN connection to but not call directly. This server is written in C++ and must...

I pushed my first cut at this in PR #918 Having a problem converting the overlapped bit count to a probability distribution. I could use some help there. The HTM...

I am still struggling with the PDF that should be returned as a result from infer( ). The PDF generated by the softmax( ) function used by SDRClassifier should be...

> I don't really understand how your solution works During learning I store only one entry for each category and ON bit combination. So, if there are 40 ON bits...

I still don't think using softmax( ) to generate the probability is getting the correct numbers but perhaps it does not matter. If the pattern being matched in the infer(...

I tried the OverlapClassifier with MNIST and the results are rather disappointing. The best I could do was 36% while the original SDRClassifier did 94.4%. I suspect ORing all training...

@aamir-gmail, I think this is basically what I did and it did not perform very well with the MNIST data. Of course I may have messed up the coding someplace....

I took another crack at implementing this Classifier using overlapping bits rather than classical AI. This time I kept track of how many training samples for a label that contained...

I did run into a problem however. If I processed the raw images it completes correctly. But when I run the data through SP, it crashes on element 1080 of...