Whether you need to swarm on each field of data depends on whether the columns of data have different characteristics. If they are all basically within the same bounds, same data type, etc., you generally won't need to swarm against all of them and you can just reuse the model params from one swarm against a field for each column. But if they are different types and have drastically different parameters, you should swarm against them all.
Since so many of you are asking about this, I guess I should put some time into finishing the "Multiple Hot Gyms Prediction Tutorial" (https://github.com/numenta/nupic/issues/1320), which would give examples of running an ensemble of models against many input fields. I started it at https://github.com/rhyolight/nupic/tree/many-hot-gyms-prediction but it is quite out-of-date at this point. --------- Matt Taylor OS Community Flag-Bearer Numenta On Thu, May 14, 2015 at 12:51 AM, email email <[email protected]> wrote: > Hi, Matt, > > I also have the same case as Tom. My question is if I create several models, > do I need to write the separate swarm description and swarm.py files for > each model and run swarm one by one? If not, how to integrate the swarm > description and swarm files, or swarm with several description configs in a > parallel form? > > Thanks > > Regards, > Richard > > On Thu, May 7, 2015 at 11:29 PM, Matthew Taylor <[email protected]> wrote: >> >> Tom, >> >> This is something we'd like to do, but doesn't work yet. >> >> https://github.com/numenta/nupic/issues/1712. >> >> Your only option at this point (and it's not a horrible option) is to >> create several models for the same input, one for each field you'd >> like to predict. Then send them all the input rows at the same time >> while they predict for their own fields. >> >> Regards, >> --------- >> Matt Taylor >> OS Community Flag-Bearer >> Numenta >> >> >> On Thu, May 7, 2015 at 12:01 AM, Tom Tan <[email protected]> wrote: >> > Hi, >> > >> > Is there an example or documentation how to predict multiple fields? I >> > looked at the example using multiple fields >> > https://github.com/subutai/nupic.subutai/tree/master/swarm_examples, and >> > it >> > is to use multiple field inputs to predict a single output. Let’s say >> > I >> > have a stream of CPU, RAM, I/O, and network data for a server. I’d >> > like to >> > CLA model to be able to predict next time period CPU, Ram, I/O and >> > network >> > values at once, and perhaps generate one anomaly score considering all >> > predictions vs. actual. Do I have to use multiple models, one for each >> > data field to accomplish this? >> > >> > Regards, >> > Tom >> > >> > >> > >> >
