Will NuPIC produce some results for those users with >500 check-ins? which is about 25 per month.
On Tue, Mar 10, 2015 at 9:53 PM, Matthew Taylor <[email protected]> wrote: > Ok, so I'm assuming a bit here, but... > > 107092 users with 6442892 total checkins over the course of (about) 20 > months.... then the average checkings per user per years is > (6442892/107092) / 20 = 3 checkins per month. > > If you want to analyze users individually with only 3 checkings per > month, that is not a very fast stream of data. I'm not sure you're > going to get anything out of it with NuPIC. > > However, if you wanted to try to analyze population trends in the > complete data set over time, it might be worthwhile to try with over > 300K total checkins per month, but you wouldn't be able to get any > analysis out of it for individual users. > > Perhaps you could group the users by geographic location to increase > the number of data points in a model if you aren't interested in > complete population analysis? > > --------- > Matt Taylor > OS Community Flag-Bearer > Numenta > > > On Tue, Mar 10, 2015 at 7:26 AM, Manal Bhingardeve > <[email protected]> wrote: > > Hi Matt, > > > > I would like to get indications of anomalous behaviour. > > The patterns should be isolated to each person. > > The data consists of check-ins of users over the period of Feb. 2009 - > Oct. > > 2010 > > These are some details about the data. > > 107092 : users > > 6442892 : check-ins > > 60.162215665 : average check-ins > > no of users : no of check-ins greater than > > 17 : >2000 > > 244 : >1500 > > 475 : >1000 > > 1488 : >500 > > > > Thanks, > > Manal > > > > On Tue, Mar 10, 2015 at 3:48 AM, Matthew Taylor <[email protected]> > wrote: > >> > >> Manal, > >> > >> Some questions... > >> > >> 1. What would you like to get out of NuPIC? Predictions of future > >> checkins? Indictions of anomalous behavior? > >> 2. Should detected patterns be isolated to each person or detected > >> from the population as a whole? > >> 3. How often do checkins occur? > >> 4. How much past data is available (for training)? > >> > >> --------- > >> Matt Taylor > >> OS Community Flag-Bearer > >> Numenta > >> > >> > >> On Mon, Mar 9, 2015 at 2:44 PM, Manal Bhingardeve > >> <[email protected]> wrote: > >> > Hi, > >> > > >> > I wanted to use CLA to find patterns in user check-ins for location > >> > based > >> > services. > >> > The data consists of check-in details for multiple users with the > >> > following > >> > fields user_id, check-in time, latitude, longitude, location id. > >> > Would the CLA be suited for this kind of an application and if so, > could > >> > I > >> > get some help with the swarming process, more specifically as to how > the > >> > geospatial encoder could be used. > >> > > >> > Thanks, > >> > Manal > >> > > > >
