Thanks Mark. See https://github.com/numenta/nupic/wiki/Using-NuPIC#predicting-sine-waves-with-nupic and https://github.com/subutai/nupic.subutai/tree/master/swarm_examples --------- Matt Taylor OS Community Flag-Bearer Numenta
On Thu, Jun 5, 2014 at 2:08 PM, Marek Otahal <[email protected]> wrote: > Hi Nokan, > > this is a regression task, as you said, the function is not temporal (no > hidden state), thus you do not need to use the temporal pooler, just > experiment with spatial pooler. > > Just train the model (SP only) on input data like this: > param1, param2, ..., paramN, resultOfFunction > same for example 2 > same for example 3 > ... > > The predict like this: > param1, param2, ..., paramN, ? > > take a look at the categorization examples here : > examples/opf/experiments/classification/category_SP_0/ > ..or even Matt's 'Sine example video tutorial' (I'm not sure if that one is > temporal, or not) > > > @Matt, it's called black-box modelling, just like your sine example, you > want to model a (complex) system based only on your available (sparse) data, > and might want to get function's results for values not covered by your > list. > > Cheers, Mark > > > > > > On Thu, Jun 5, 2014 at 4:14 PM, Matthew Taylor <[email protected]> wrote: >> >> Dear U, >> >> If there is no temporal component to your problem, I'm not sure NuPIC >> will work for you. But I don't really understand the problem. Are you >> trying to predict the outcome of a function based upon the input >> params of the function? Silly question, but why not just run the >> function instead of a prediction engine? >> --------- >> Matt Taylor >> OS Community Flag-Bearer >> Numenta >> >> >> On Thu, Jun 5, 2014 at 6:15 AM, Nokan Emiro <[email protected]> wrote: >> > Hi, >> > >> > I'm a newbie here, so please forgive me for this silly questin. >> > >> > My original problem that brought me here is that I need a solution that >> > is able to learn by examples, how to calculate a pure function. I have >> > a >> > fair amount of examples that can be used to teach the thing, and my >> > expectation is that it should be able to approximate the whole function >> > even with parameters never seen before. I thought NuPIC is something >> > that's capable of this, but the more I read about how it works, the less >> > I'm sure about it. >> > >> > The function is a pure function in terms of functional programming. It >> > does not have any internal state, no side-effects. It just calculates a >> > return value from its params. No temporality is involved, there's no >> > context of the calculation other than what's encoded in the params. The >> > function operates on huge integer numbers and the result of the >> > calculation >> > is also a huge integer. I guess I can use a Scalar encoder to put the >> > numbers into the "brain", but how can I get the result? maybe I should >> > provide the input and the output both as inputs at the same time, and >> > somehow use nupic to "predict" the output part when the input part is >> > given? >> > >> > Can you please tell me if nupic is something capable of that? And if >> > it is, where can I find examples of documentations that help me start >> > playing with it as fast as possible? >> > >> > THX, >> > U. >> > >> > >> > >> > _______________________________________________ >> > 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 > > > > > -- > Marek Otahal :o) > > _______________________________________________ > 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
