Hi, Just wanted to add a couple more notes:
1) The swarm gives you a report indicating the contributions from each field. You can use that to double check that it did run. Look for the line "Field Contributions” near the end of the console output. A positive value indicates adding the field reduced the error. 2) A large swarm will try more field combinations than a medium swarm, so you might want to try that. It will take a lot longer though. 3) You can actually force a specific field combination, using the “fixedFields” attribute. So you can add something like: "fixedFields": [“AWT_01”,”Tilt_01"], This will force swarm to only use the above two fields. It won’t try any other field combinations, but it will optimize the other parameters. This is much faster since it has many fewer experiments to run. Note that temperature may not always help. Just because a field is statically correlated with the predicted field does not mean it helps in making predictions. In fact, the more correlated a field is with the predicted field, the more likely that it is unnecessary and will be left out. (This is the opposite of what happens in most machine learning applications.) --Subutai On Thu, Aug 21, 2014 at 10:52 AM, Matthew Taylor <[email protected]> wrote: > No, you just have to add it afterwards. Are you certain that AWT_01 > will contribute to the Tilt_01 predictions? The swarm seems to think > otherwise. You can investigate the swarm results because it leaves > behind a folder with all the details from the swarms. You might look > inside to check and see what models it created and what error scores > were returned to prove to yourself that the swarm include a model with > AWT_01 encoders. > > If you provide your data file, I might find time to run some tests and > help out. > --------- > Matt Taylor > OS Community Flag-Bearer > Numenta > > > On Thu, Aug 21, 2014 at 8:45 AM, Cavan Day-Lewis > <[email protected]> wrote: > > Classification: NPL Management Ltd - Commercial > > > > I have manually added an encoder in model params for the temperature > > field, but this is not ideal. Is there are way to force the swarm to > > find an encoder for the temperature field? > > > > Thanks, > > Cavan > > > > -----Original Message----- > > From: nupic [mailto:[email protected]] On Behalf Of Cavan > > Day-Lewis > > Sent: 21 August 2014 10:24 > > To: NuPIC general mailing list. > > Subject: Re: [nupic-discuss] Prediction of multiple time series for > > concrete bridge problem [NC] > > > > Classification: NPL Management Ltd - Commercial > > > > Thanks Matt, > > > > I ran a medium size swarm, over a larger portion of the data > > (iterationCount: 10000) but it still didn't include an encoder for the > > temperature field in the model params. > > > > Cavan > > > > -----Original Message----- > > From: nupic [mailto:[email protected]] On Behalf Of > > Matthew Taylor > > Sent: 20 August 2014 22:14 > > To: NuPIC general mailing list. > > Subject: Re: [nupic-discuss] Prediction of multiple time series for > > concrete bridge problem [NI] > > > > In the swarm_description.py, your "swarmSize" is set to "small", which > > won't give you good results. I suggest you set it to "medium". This > > takes much longer, but the "small" size is just for debugging. This > > might be the source of all your problems. > > > > If not, and you just accidentally left it at "small" while debugging... > > > > On Wed, Aug 20, 2014 at 1:58 AM, Cavan Day-Lewis > > <[email protected]> wrote: > >> Note: in TimeTempTilt_model_params.py it generated: > >> > >> > >>> 'sensorParams': { 'encoders': { u'AWT_01': None, > > > > You have the swarm configured to optimize for predicting "Tilt_01". It > > seems that the swarm did not find that the "AWT_01" value contributed to > > better prediction of "Tile_01", and that's why it didn't include an > > encoder for it. > > > > Hope that helps, > > --------- > > Matt Taylor > > OS Community Flag-Bearer > > Numenta > > > > _______________________________________________ > > nupic mailing list > > [email protected] > > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > > > -- > > If you have received this message in error, please notify us and remove > > it from your system. > > NPL Management Ltd cannot guarantee that the e-mail or any attachments > > are free from viruses. > > > > NPL Management Ltd is a company registered in England and Wales, number: > > 2937881 Registered office: Serco House | 16 Bartley Wood Business Park | > > Hook, Hampshire | UK | RG27 9UY > > > > > > _______________________________________________ > > 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 > > _______________________________________________ > 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
