On 11/06/2015 12:12 PM, Jos Theelen wrote:
Is it possible to use the predictive cells for prediction?
I have a model with a limited number of future events I want to predict.
So I calculated for each event the SDR with the encoder and spatial
pooler. Next I compared those SDR's with the predictive cells of the
temporal memory. When the overlap was big, I assumed that it was a
likely event. When the overlap was small or zero, I assumed that the
future event almost certainly wouldn't happen.
Is that a good method? Has someone else done this before? Is it correct
to say that if the overlap is bigger, then that event is more likely to
happen?
This is very interesting can you please provide some code or brief
tutorial ... e.g. which API are you using etc? I am asking because so
far I've used OPF and I was able only to get predictions and anomaly
score and I always wanted to know ho to do such low level stuff with
NuPIC and this seems like interesting idea.
Thank you very much
Wakan Tanka
