Yes, the same parameter is used in both conditions. If a segment already exists then we first subtract out the number of synapses that are already active. There is also a general upper limit on the total number of synapses that can be on a segment (maxSynapsesPerSegment).
There is definitely an interaction between newSynapseCount and the thresholds. We usually keep newSynapseCount fixed, and adjust the thresholds based on the noise in each dataset (this is done automatically in the swarm process). --Subutai On Wed, Jul 24, 2013 at 3:54 PM, Ryan J. McCall <[email protected]> wrote: > Hi Subutai, > > Okay, now the whitepaper says that 'newSynapseCount' defines "the maximum > number of synapses added to a segment during learning" so does it follow > that > 'newSynapseCount' defines both the initial synapses per segment and the > max number of synapses added when performing a segment update? > > Thank you, > > Ryan > > > On 7/24/13 5:47 PM, Subutai Ahmad wrote: > > Hi Ryan, > > That's right, it needs to be greater than minThreshold. There is a > variable called newSynapseCount in there that controls this parameter. > > Cheers, > > --Subutai > > > On Wed, Jul 24, 2013 at 3:39 PM, Ryan J. McCall <[email protected]>wrote: > >> Hi all, >> >> I was studying the temporal pooler parameters used in the file: >> >> /examples/opf/experiments/opfrunexperiment_test/simpleOPF/hotgym/description.py >> >> I'm curious as to how many synapses should be initially created per >> distal dendrite segment. I didn't find a parameter defining this. Based on >> my understanding from the whitepaper, it seems that there has to be at >> least 'minThreshold' number of previous active synapses on a dendrite >> segment to have any sequence segment updates queued. >> >> # Minimum number of active synapses for a segment to be considered >> # during search for the best-matching segments. >> # None=use default >> # Replaces: tpMinThreshold >> 'minThreshold': 12, >> >> So is there a guideline on how many synapses distal dendrite segments >> should be created initially? For instance, if 'minThreshold' is 12 and we >> only have 11 initial synapses per segment it appears to me that it would be >> impossible for the temporal pooler to ever "get off the ground" so to >> speak. My guess is that there would need to be at least x2 'minThreshold' >> initially... Many thanks, Ryan McCall >> >> -- >> Ryan J. McCall >> Ph.D. Student, Dept. of Computer Science >> Cognitive Computing Research Group >> Institute for Intelligent Systems >> The University of Memphis >> ryanjmccall.com >> >> _______________________________________________ >> nupic mailing list >> [email protected] >> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >> >> > Take Survey & Get jetblue 1000 visa Limited Time Offer for US Only ! > > http://click.lavabit.com/widzzh7qp1hw3q5mkouoh8cnizb1s4am9876z4dmukojcj68s4oy/ > > > _______________________________________________ > nupic mailing > [email protected]http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > > > -- > Ryan J. McCall > Ph.D. Student, Dept. of Computer Science > Cognitive Computing Research Group > Institute for Intelligent Systems > The University of Memphis > ryanjmccall.com > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
