Julie, I don't see anything obvious you are doing wrong. Would it be possible to send a sample data file that we can use to try and replicate the error? (e.g. just put in a sample session where you are going through numenta.orgor something). A file with a few dozen or hundred rows should be fine.
11 fields should be fine, though it does take a while for the swarm to go through all those combinations. As far as URL and predicted field, the OPF always tries to predict the value at the *next* time step. It does not try to predict URL from the other values in the same row, but instead tries to use all the rows that preceded it including previous values of the URL field. This is a key difference with other machine learning techniques which tend to be non-temporal in nature. --Subutai On Fri, Apr 11, 2014 at 11:35 AM, Julie Pitt <[email protected]> wrote: > Same result. :( I did have to back out of build/scripts in order to run > make. > > > On Fri, Apr 11, 2014 at 9:55 AM, Matthew Taylor <[email protected]> wrote: > >> mkdir -p $NUPIC/build/scripts >> cd $NUPIC/build/scripts >> cmake $NUPIC >> make -j2 >> >> If you don't run cmake from within the "build/scripts" directory, I'm >> not sure it will build correctly. >> --------- >> Matt Taylor >> OS Community Flag-Bearer >> Numenta >> >> >> On Fri, Apr 11, 2014 at 9:50 AM, Julie Pitt <[email protected]> wrote: >> > Are these the correct commands to re-build on mac? >> > >> >> cmake $NUPIC >> >> make -j2 >> > >> > I just pulled nupic and ran the above commands in the nupic/ directory. >> I >> > tried the run swarm again and got the exact same error. >> > >> > >> > On Thu, Apr 10, 2014 at 5:59 PM, Julie Pitt <[email protected]> wrote: >> >> >> >> Will do. I violated rule #1: always check out and build the latest code >> >> before complaining :) >> >> >> >> If anyone has answers to my 3 questions below, that would be helpful. >> >> >> >> >> >> On Thursday, April 10, 2014, Matthew Taylor <[email protected]> wrote: >> >>> >> >>> Julie, I recently fixed a bug in swarming: >> >>> >> >>> https://github.com/numenta/nupic/pull/793 >> >>> >> >>> Please make sure your codebase has this fix and try again? >> >>> --------- >> >>> Matt Taylor >> >>> OS Community Flag-Bearer >> >>> Numenta >> >>> >> >>> >> >>> On Thu, Apr 10, 2014 at 4:10 PM, Julie Pitt <[email protected]> >> wrote: >> >>> > I am probably missing something obvious (working on the URL >> example). I >> >>> > am >> >>> > trying to swarm over my browsing history. The search_def is here. >> >>> > >> >>> > Sorry I didn't include my data file, but I don't feel like posting >> my >> >>> > browsing history on the internet. If you're curious, you can harvest >> >>> > the >> >>> > data yourself if you use Chrome/Mac by following the README. I get >> this >> >>> > error when I try to swarm: >> >>> > >> >>> > Traceback (most recent call last): >> >>> > File "/Users/julie/github/nupic/nta/eng/bin/run_swarm.py", line >> 29, >> >>> > in >> >>> > <module> >> >>> > permutations_runner.main() >> >>> > File >> >>> > >> >>> > >> "/Users/julie/github/nupic/nta/eng/lib/python2.7/site-packages/nupic/swarming/permutations_runner.py", >> >>> > line 126, in main >> >>> > return runPermutations(sys.argv[1:]) >> >>> > File >> >>> > >> >>> > >> "/Users/julie/github/nupic/nta/eng/lib/python2.7/site-packages/nupic/swarming/permutations_runner.py", >> >>> > line 418, in runPermutations >> >>> > metricsKeys=search.getDiscoveredMetricsKeys()) >> >>> > File >> >>> > >> >>> > >> "/Users/julie/github/nupic/nta/eng/lib/python2.7/site-packages/nupic/swarming/permutations_runner.py", >> >>> > line 902, in generateReport >> >>> > bestModelNum = results['bestModel'] >> >>> > KeyError: 'bestModel' >> >>> > >> >>> > Is there something obvious I'm doing wrong? >> >>> > >> >>> > Other questions I have are: >> >>> > >> >>> > 1. There are 11 fields in the swarm. Is that too much? >> >>> > 2. In many cases, there is no value for one or more fields (e.g., >> there >> >>> > is >> >>> > no subdomain, or if the URL has an IP, there is no TLD). Is this a >> >>> > problem? >> >>> > 3. I want to predict the URL from the other fields, so I made 'url' >> the >> >>> > predicted field. Unfortunately I can't make it a predicted field >> >>> > without >> >>> > also making it an input to the model. Is that expected? I'm >> wondering >> >>> > if the >> >>> > presence of the entire URL in the SDR will mess up the prediction. >> >>> > >> >>> > >> >>> > >> >>> > >> >>> > _______________________________________________ >> >>> > 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 >> > > > _______________________________________________ > 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
