Karim, Sorry, I should've been more clear. My answer applies to NuPIC as well. It's more suited for prediction and anomaly detection of sequences it has already learned, specifically predicting the elements in the sequence.
- Chetan > On Jul 24, 2015, at 9:25 PM, Karim <[email protected]> wrote: > > Hi Chetan, > I wanted to know if problem can be solved by Nupic. Not necessarily by the > geospatial example. > >> On Saturday, July 25, 2015, Chetan Surpur <[email protected]> wrote: >> Hi Karim, >> >> Answers below: >> >>> On Jul 24, 2015, at 4:14 AM, Karim <[email protected]> wrote: >>> >>> Hello, >>> >>> I have GPX trails of vehicles/bikes of a particular city. Meaning, lat/lon, >>> timestamp of bike trails from where a particular bike started and ended at >>> different hours of the day. I have sets of these trails for different bikes >>> roaming around the city. >>> >> >> Great! This sounds like a good format for feeding into NuPIC using the >> geospatial encoder. >> >>> >>> I have seen anomaly detection video of NuPic and liked it. Now, I want to >>> know if it is possible with NuPic to know: >>> >>> 1. Given new two random points A and B, how much time it takes to go from >>> point A to point B in a particular range of hours of the day by using the >>> GPX data I already have. Assume I have enough data for the city in terms of >>> bike trails. >>> >> >> Unfortunately, this is not a problem well-suited for the geospatial demo in >> NuPIC currently. The kinds of problems that are suited for it are anomaly >> detection and prediction of already-seen trails. For example, detecting >> whether a bike on its usual trail stops suddenly or goes off-track. >> >> - Chetan
