TODO

TODO:

(1) Input routes for all NYC buses -- or at least the one we're tracking
(2) Geocode and reverse geocode all locations
  (a) geocoding for input can use google -- DONE
  (b) geocoding for output can use street information from tiger:
    (i) a route is an ordered set of street segments
    (ii) we localize to which segment we're on, using min(st_distance(...)) and 
         st_dwithin for optimization.  preprocess data with st_linemerge.
    (iii) that gives us a gid.  For the "and", we figure out
          what end we're closest to, and we use st_intersects with st_startpoint
          or st_endpoint to find the cross-street.  
    (iv) Or we could just use a web service. -- DONE

(3) represent routes as a set of route segments. -- DONE

(a) this allows use of unmodified tiger data -- DONE

(b) it allows calculating number of intersections, by doing:
count(*) where st_intersects() on the subset of the route between here
and there.  I guess "between" means "compute the segment that we're on",
"compute the segment the target is on", and use the route's segment
ordering.

(4) using the number of intersections (DOne0 and bus stops to better
estimate times.

(5) warn when bus data is very out-of-date

(6) decide on normalized vs denormalized geometry for routes. -- DONE, chose normalized.

normalized: cleaner
denormalized: many route calculations more efficient; we have to do work
to define routes anyway.

(7) various testing:

(a) given a set of observations, generate estimates as-of various time
points.

(b) how accurate are the estimates?

(c) are paths correct?

(8) Demo: gather real data for M6 bus, at least one run.  Show route
on a map with time points.  Have a slider representing query time, 
which shows only data points as-of that time.  A dropdown list of bus stops.

Select a bus stop and a query time; the bus stop shows up on the map, and
the time estimate and true arrival time are shown.  

(9) Improve estimate of arrival time.  Use neural nets.

(10) test is sensitive to number of points; fix this -- no it's not (DONE).

(11) only collect two times per points: arrival, departure -- DONE

(12) provide map-matched points via kml

(13) FIX: why are sql commands at end of management commands not executed? -- DONE