alanlujan91 / redistricter Goto Github PK

Dropping all the state lines, split the USA into 50 or 100 or 435 districts.

The hard part: connecting geographic data of adjoining states. It'll be an 
annoying merge operation to find all the matching tiny edge pieces. Then, write 
out one giant merged geometry file and node adjacency and so on.

http://maps.stamen.com/
http://openlayers.org/
http://code.google.com/apis/maps/documentation/javascript/
http://openstreetmap.org/

g++ -g -Wall -Itiger -MMD fileio.o Bitmap.o tiger/mmaped.o Solver.o
District2.o PreThread.o renderDistricts.o LinearInterpolate.o
GrabIntermediateStorage.o AbstractDistrict.o DistrictSet.o
NearestNeighborDistrictSet.o rtaToDsz.o tiger/recordA.o -lpthread -lpng12
-lz -o rta2dsz
make: *** No rule to make target `uf1.o', needed by `uf1data'.  Stop.

Districts only use 1 byte, allowing for 254 or so of them. Use two bytes, allow 
New Hampshire to work.

Blocks connected by roads are better than blocks that just happen to be 
adjacent. Maybe most noticable around rivers or coastal islands. 
Water-discontiguity is probably available in the data already processed and 
road-contiguity is probably available in Census map files.

Improvement: warp latitudes and longitudes onto a polar projection centered 
over the center of the state. Regularize dimensions to X,Y integer millimeters 
around some 0,0 center.
This will improve dimensional accuracy in all states, more so the northern 
ones, and should fix Alaska's -180/180 wraparound problem.

Alaska is broken. Treating lon/lat as X/Y really doesn't work there. Find a 
center lat-lon and do a polar projection that mostly preserves linear distances.
Then, do that for all the states.

Find sources of precinct level election results data. Translate precincts to 
their census blocks. Distribute votes per census block. Sum up per-census-block 
votes into new map block assignments.

Create 'keep together regions' which, if not wholly in one district, the 
population in that district counts as some adjustment further away from the 
center of their districts.
 (auto populate these from census city and metro-area and county regions)
Create 'do not cross lines' which, if they bisect a district, the population on 
the far side of the line counts as some adjustment further away from the center 
of the district.