On Sun, Jun 14, 2015 at 5:12 PM, Chris Bennight <[email protected]> wrote:
>
>
> I also went ahead and rendered a graphic of what they look like
>
> Excellent!  In fact, this immediately reveals its value, because I see now
> that the query polygon is a random star polygon with significant
> concavity.  Definitely better than a regular convex polygon!  8^)
>
> Re: Synthetic vs. natural
> Completely reasonable.  Probably using variable simplified coastlines
> would be a good real world case analogue (with a randomly generated query
> geometry - asking the question "what countries does this query window
> overlap).
>

Um, maybe.  All depends on use case.  There's a pretty much infinite number
of scenarios.  Here's some (invented) others:
- What US census blocks intersect some (larger / smaller) query area?
- What ship/truck tracks intersect some area of interest?

>
>
> Re: highly data-dependent
> That makes sense - what I"m hopefully trying to come up with are a set of
> heuristics that I can use to determine which strategy to use (with the
> caveat that I may have to collect stats/profile a data set, and it's not
> going to necessarily be correct for every comparison - just hoping to
> answer the "best default setting")
>

In absence of *any* knowledge of the characteristics of query and target, I
think the heuristic is  "use PreparedGeometry".  Maybe there's a refinement
to use raw Geometry if the query polygon is low vertex count, or if the
target set size is small.   And if you find a systemic performance issue
with very large query PGs, that's definitely of interest.

>
> Re: Increasing vertex count by densifying
> I think taken in reverse this might be a good analogue for expected
> simplification gains?  (Granted the is also a decreasing amount of
> irregularity in a real simplification, but unless intersects optimizes for
> co-linear points I expect the cost is the same).
>

Lower vertex count is always good  8^)  Intersects does not optimize for
collinear points.  In fact, using monotone chains actually penalizes
collinear vertices.  (An interesting question is whether adding a maximum
run length to MonotoneChain construction would improve performance.  But
this seems to me like a highly atypical case).
------------------------------------------------------------------------------
_______________________________________________
Jts-topo-suite-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/jts-topo-suite-user

Reply via email to