Thanks, much better! After adjusting Scale too (5m default is too small when data points are taken so infrequently (one per minute), 25m was good ), all expected effects of learning appeared.
Kalle T. -----Original Message----- From: nupic [mailto:[email protected]] On Behalf Of Matthew Taylor Sent: Wednesday, March 25, 2015 5:29 AM To: Kalle Tammemäe Subject: Re: Experience with NuPIC Geospatial tracking using Android GPSLogger There is an option to the run.py script (-m) that allows manual sequences based upon the "tracks" defined in the GPX file. https://github.com/numenta/nupic.geospatial/blob/master/run.py#L43 --------- Matt Taylor OS Community Flag-Bearer Numenta On Mon, Mar 23, 2015 at 8:36 AM, Kalle Tammemäe <[email protected]> wrote: > Hi, > > > > Does someone have experience with using of GPS data recorded using > GPSLogger (version 59), on Android (5.0.2) phone? > > Namely, the Geospatial tutorial data seems to be recorded with > relatively high sampling rate (a sample per second), GPSLogger is > recording one sample per minute. > > Due to that preprocessing of csv data is failing, rows are discarded. > > > > It can be corrected commenting out line in tools/preprocess_data.py: > “if (lastTimestamp and (timestamp - lastTimestamp).total_seconds() > > 5): keep = False” > > After that most of samples will be accepted. > > > > Nevertheless recognition the track sequences during run.py is > obviously failing, system seems to interpret every data row as a new sequence. > Therefore there is no learning either anomaly detection. > > On map I see track threads (I had set of similar walks), symbols of > sample points all oriented to NW direction. > > > > Regards, > > Kalle Tammemae > >
