Marcus Lewis

Oops, Github converted my hyperlink to a cross-reference, and I have no way to delete it. Thanks, the code in the preceding comment solves this for me.

> RandomImpl->getUInt32(void) does not do what one'd expect (uniform random numbers 0..UInt32:max), the samples just a HALF Nice catch! This means that the random number generator is biased in the...

I think this issue is just saying "The encoders' `getOutputWidth` method should return a `UInt`, not an `int`". That sounds reasonable. I'd support that change, but I think https://github.com/numenta/nupic.core/pull/962 is...

You're correct, this could be better. The issue doesn't currently affect TemporalMemory, since there is never more than one destroyed segment. But it's pretty easy to exhibit the issue with...

The CellData is not part of the public interface of the Connections. I'm not sure we should support extending it. It's an internal detail that we should be able to...

Templates are definitely an option. And it would be nice to have our encoder output type be more compatible with our SpatialPooler input type (`UInt` array of 0s and 1s)....

I agree this change would make sense. This `newSynapseCount` (TP) / `maxNewSynapseCount` (TM) parameter has always seemed a little goofy. It's actually treated as a "desired active synapse count". So...

With this philosophy of "Segments should keep adding synapses", there's other Temporal Memory behavior that now seems wrong. In the "activate predicted column" path, synapses are never grown. We only...

@subutai Status update: I've made both of these changes, and I've found that: - Enabling synapse growth in `activatePredictedColumn` doesn't break anything, aside from some unit tests that were verifying...

Do we know whether the results for `tm_py` are as good as `cpp`? I can definitely run it overnight and check. Really we just want the NAB results to be...