Thanks Fergal :) When we combine 80 separate variables, everything part of single sliding window for the encoder to bring better overlapping? or each variable encoded separately and combined and feed into SP? In my testing, if i arrange my data sequentially then it makes the better prediction than spatial arrangement.
Thanks Ramesh Ganesan. On Fri, Aug 9, 2013 at 4:54 PM, Fergal Byrne <[email protected]>wrote: > > Hi Ramesh, > > Well, if you're encoding each one with 128 bits, you're trying to get > one region to learn sequences of 80 separate variables (80x128 = 10k). If > there is significant (but possibly invisible or unexplained) semantic > correlation among your 80 variables, then a 2048-column region may well be > able to handle it, because the correlated inputs will have some synchrony > and will appear as a combined value for each record. > > If the variables are uncorrelated, you're living in an 80-dimensional > data space, so good luck with that! > > Regards, > > Fergal Byrne > > > > > > _______________________________________________ > 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
