A short answer for now, hopefully I can give a longer response tomorrow from the office.
First of all, if you haven't read these docs, read them: - https://github.com/numenta/nupic/wiki/Online-Prediction-Framework - https://github.com/numenta/nupic/wiki/Using-NuPIC Specifically, there is a section at https://github.com/numenta/nupic/wiki/Online-Prediction-Framework#model-output-the-modelresult-object that talks about the model results you are asking about. 2nd, we have a search page at http://numenta.org/search.html?q=modelresults that might help you serve yourself as you have further questions. It indexes our source code, wiki, website, and mailing list archives. Regards, --------- Matt Taylor OS Community Flag-Bearer Numenta On Sun, May 17, 2015 at 6:04 PM, Wakan Tanka <[email protected]> wrote: > Hello guys, > I've been searching in mailing list but did not get the answer, I'm not sure > if this should go here or [email protected] because it overlaps > with theory. I decided to post here because it is probably more practical. > Can anybody explain the csv file which is generated by swarm proces + > running model? I am using tutorial from here: > > https://www.youtube.com/watch?v=KuFfm3ncEwI > https://github.com/rhyolight/nupic.examples/tree/master/sine-prediction > > and my csv file is looks like this: > http://termbin.com/k4pu > > > > CSV Fields: > ########### > > # Original values, specific to this case > # why is there string instead of float > angle > sine > > # what is this? > multiStepPredictions.actual > > # What is this? > # Does it contains two values separated by : (for angle and sin) ? > multiStepPredictions.1 > > # Is this field some type of score that is > # assigned based on what was expected an what really comes? > # Does it contains two values separated by : (for angle and sin) ? > anomalyScore > > # What is this? > multiStepBestPredictions.actual > > > # From the mentioned video I've figured out that this is the best > # prediction (chosen from multiple that was generated by swarming) > # for one step. > # How This is related to anomaly score? > multiStepBestPredictions.1 > > # This field has type "string" and it contains records like: > # [] > # 0.0314107590781 > # 0.21814324139654254: 0.043062200956937892 > anomalyLabel > > # This field has type "float" and it contains the same records > # types like anomalyLabel. So why it is once string and second float? > # I think this is specific to this case > multiStepBestPredictions:multiStep:errorMetric='altMAPE':steps=[1]:window=1000:field=sine > > # I think this is specific to this case > multiStepBestPredictions:multiStep:errorMetric='aae':steps=[1]:window=1000:field=sine > > > What about the other columns that does not have header? > > > Here is described the NuPIC Input Data File Format. There is said that it is > required for running swarms and OPF but I assume the actual output of > swarming + running model gives the same format (three first rows == header) > https://github.com/numenta/nupic/wiki/NuPIC-Input-Data-File-Format > > > > PS: the fields of csv are optional or they are hard coded for every csv > being generated e.g. is possible to change order or omit some fields? > > PPS: I'm viewing csv via following command I hope the columns corresponds to > types as I described (it looks similar to office from mentioned video, but I > prefer command line): > column -s,$'\t' -t < *.csv | less -#2 -N -S > > PPPS: Are those fields somehow related e.g. the score is computed as some > inputs from previous step or something like that? > > > > Thank you > > Regards > > > Marek >
