prediction quality using ApicalTiebreakTemporalMemory?

[Thanh-Binh]

Dear all, I am using 2 versions of ApicalTiebreakTemporalMemory

1. very old C++ version in htmresearch.core of Numenta (very complicate codes)
2. very new Python version in advanced folder of htm.core (very understandable and clean codes) for predicting a simple wave like sin() (exactly: ApicalTiebreakSequenceMemory). The old version functions well and can predict after very short learning.

However, the new one in htm.core can not predict it even after very long learning time. Does anybody test it any have idea to fix the bugs? Thanks

[Thanh-Binh]

I found partly that the option maxSynapsesPerSegment = -1 causes it fail to predict. By setting: maxSynapsesPerSegment = 255, it works.

But I am not sure why? Maybe due to the fact the number of synapses per segments increases dramatically??? Does anyone have idea?