Spatial Pooler: investigate using Connections as backend about htm.core HOT 6 CLOSED

I'd very much prefer to use Connections over SparseMatrix. Connections is custom tailored for the job of running synapses. It appears to do its job well, and the source code is legible.

from htm.core.

If its possible to add methods to the Connections class then I'd recommend moving the method "adaptSynases" from the SP to the Connections. This method does Hebbian learning which is used by both the SP & TM. It could also be useful for other experiments in the future?

from htm.core.

moving the method "adaptSynases" from the SP to the Connections. This method does Hebbian learning which is used by both the SP & TM. It could also be useful for other experiments in the future?

It definitely is, I like the idea. Endgoal is for SP/TM both running connections

from htm.core.

I've started looking at this task and what things this will change:

• Move raisePermanencesToThreshold to connections?
• Move bumpSegment to connections?
• Remove synPermTrimThreshold? This has public setter/getters but I suspect that they exist to allow testing.
• Move connectedCount to connections. It has a public getter method in SP. Should connections add bookkeeping for this or can it just count synapses when called? Connections does not currently keep this information. SP does not currently use this information.

I've started work at: https://github.com/ctrl-z-9000-times/nupic.cpp/tree/connections but it does not yet compile.

from htm.core.

Connections does not currently keep this information. SP does not currently use this information.

ConnectedCount ..of nobody uses it, just rm

from htm.core.

Successfully implemented in #153

from htm.core.