Hi Valentin, maybe I misunderstood your question, so let me check.. what kind of data are you streaming? If you say you have a SDR data, that means it's a SDR at a time, so you can feed an array of 1s and 0s to the TP. That will be data at time T, no prediction too far in future. From what you describe later, it looks you have a stream of 1/0, that observed for a longer time, looks like a SDR. In this case, it's like you have scalar data, you need to use an encoder, SP and feed to TP, is this the case?
Sorry, I got confused here :) Cheers, Mark On Wed, Aug 7, 2013 at 5:03 PM, Valentin Puente <[email protected]> wrote: > Hi Patrick, > > Yes the input data is flowing at the same rate of the prediction. The > number of bits is "flexible" since each single bit has full semantic > meaning (both for the input and the prediction). I dont know if feeding the > data in the TP will be good idea (it will require a very large chunk of > data to be fee.. and probably the prediction could too far in the > future.... ). Probably, i could need something smaller and the the SP will > be required. We will see... :-) > > > > > 2013/8/6 Patrick Higgins <[email protected]> > >> Hi Valentin, >> >> 1. How many bits long is one set of data from your >> source? What I mean by 'one set' is a complete >> group or record of 0s and 1s from each of the >> independent places that report an output before >> the sequence repeats. >> >> 2. How quickly will each record become available? >> >> >> I'm hearing you say you want to feed the data >> into the TP directly, but I don't see how that can >> work. You'll need to send it thru the front end of >> the CLA so it can first run your data thru the SP >> first and those patterns will then be fed to the TP. >> >> Your data sounds sparse enough for the CLA >> to work with. I'd start by feeding your raw data >> into the CLA as records and experiment with >> the setting until you find its able to learn the >> patterns and accurately predict the next record. >> If your data has semantic meaning and you have >> a fast paced changing set of data, then the CLA >> should be able to find patterns and make predictions. >> >> >> >> >> Patrick >> >> >> >> >> >> On Aug 6, 2013, at 2:08 PM, Valentin Puente wrote: >> >> > That was my idea: the encoding it is a complex task to it in HW. I >> think the TP and SP are doable (especially if you use non-volatile memory >> devices). >> > >> > My approach was to use the examples of the temporal pooler as start >> point (nupic/examples/tp/). Do you think that it could easier to start from >> the "real thing" and just bypass internally the encoding? I'm a bit lost >> here :-) >> > >> >> >> _______________________________________________ >> nupic mailing list >> [email protected] >> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >> > > > > -- > -- > vpuente > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > -- Marek Otahal :o)
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
