Hi Mike, Each prediction step is calculated by the CLA Classifier using a separate probability table for that step. So the more steps you have, the more probability tables you have, the more memory it takes.
Hint: You can specify a step list like '10, 20, 30, 40, 50, 60' to get predictions for every 10 steps up to 60. This might reduce your memory requirement while still giving you useful predictions. - Chetan On Mon, Mar 17, 2014 at 6:25 AM, Mihai Polceanu <[email protected]> wrote: > Hello Chetan, > > Thank you for the hints, I managed to extract the prediction sequence for > 60 steps ahead using a "steps" list like '1, ..., 60'. > > It does seem to use a lot of memory for it, over 900mb for 60 steps and > over 300mb for 10 steps, can't tell if it's some kind of memory leak or > it's just the way it works. > > Also, it is interesting to note that the prediction sequence looks quite > well after some time (required to learn), however it does "shake up" a > little from time to time. > > Best regards, > Mike > > P.S. I hope this email is correctly interpreted as a reply to the former > one, as I added the "Re:" to the subject :-) sorry if it's not. > > _______________________________________________ > 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
