Hi Kyle, I'd love to make Shapely faster, but I'm regretting that I made coordinate sequences mutable. Also, the underlying GEOS pointer is an implementation detail that might disappear in some future version.
Can you post your profile stats? On Wed, Jan 26, 2011 at 1:05 PM, Kyle Cronan <[email protected]> wrote: > Hi all, > > I think I found what might be a bug while trying to optimize some code > that uses Shapely. In my program's innermost loop I create a Polygon > object and pass it to the 'intersects' method of another geometry. > When I profiled my code, I found that it was spending a lot of time in > Polygon.__init__, so I've been looking for a way to just set the > coordinates that make up the polygon instead of recreating the object > each time. Which would make sense because the polygons in this inner > loop always have the same dimensionality and number of vertices- only > the coordinate values change. > > Anyway, I'm using Shapely 1.2.7. I'm wondering why it doesn't work to > do something like this: > >>>> p1=Polygon(((-1, -1), (1, -1), (0, 0))) >>>> p2=Polygon(((-1, 1), (1, 1), (1, 2), (-1, 2))) >>>> p2.intersects(p1) > False >>>> list(p1.exterior.coords) > [(-1.0, -1.0), (1.0, -1.0), (0.0, 0.0), (-1.0, -1.0)] >>>> cs=lgeos.GEOSGeom_getCoordSeq(p1.exterior.__geom__) >>>> lgeos.GEOSCoordSeq_setX(cs, 2, 0) > 1 >>>> lgeos.GEOSCoordSeq_setY(cs, 2, 2) > 1 >>>> list(p1.exterior.coords) > [(-1.0, -1.0), (1.0, -1.0), (0.0, 2.0), (-1.0, -1.0)] # okay, looks > like that worked >>>> p2.intersects(p1) > False # but this is not the expected result >>>> p1 = Polygon(p1.exterior) # rebuild the polygon >>>> p2.intersects(p1) # you sure about that? > True # ah ha! > > I actually got this CoordSeq_setX/Y trick to work in an older version > of Shapely (1.2.3), but I found it would stop working the first time I > put a new object reference in p1, or if I fork()ed! Which seems > bizarre. In both cases the libgeos was 3.2.2. > > I realize I'm messing around with interfaces I probably shouldn't be, > so forgive me for asking for help with a nasty hack! But when I did > have it working with the older Shapely version, it resulted in a very > significant speedup for my application. So if anyone knows what > causes this or maybe a better way to achieve the same result, I'd > really appreciate the advice. > > Thanks, > Kyle Cronan > _______________________________________________ > Community mailing list > [email protected] > http://lists.gispython.org/mailman/listinfo/community > -- Sean _______________________________________________ Community mailing list [email protected] http://lists.gispython.org/mailman/listinfo/community
