[R] finding euclidean proximate points in two datasets

[David Winsemius]

On May 20, 2010, at 10:02 AM, Alexander Shenkin wrote:

> Hello all,
>
> I've been pouring through the various spatial packages, but haven't  
> come
> across the right thing yet.

There is a SIG for such questions.

>
> Given a set of points in 2-d space X, i'm trying to find the subset of
> points in Y proximate to each point in X.  Furthermore, the proximity
> threshold of each point in X differs (X$threshold).  I've constructed
> this myself already, but it's horrificly slow with a dataset of 40k+
> points in one set, and a 700 in the other.
>
> A very inefficient example of what I'm looking for:

Not really a reproducible example. If euclidean_dist is a function ,  
then it is not one in any of the packages I have installed.

>
>    for (pt in X$idx) {
>        proximity[i] = euclidian_dist(X[pt]$x, X[pt]$y, Y$x, Y$y) <
> X$threshold
> 	i = i+1
>    }
>

Have you considered first creating a subset of candidate points that  
are within "threshold" of each reference point on both coordinates.  
That might sidestep a lot of calculations on points that are easily  
eliminated on a single comparison. Then you could calculate distances  
within that surviving subset of points. On average that should give  
you an over 50% "hit rate":

 > (4/3)*pi*0.5^3
[1] 0.5235988



> Perhaps crossdist() in spatstat is what I should use, and then code a
> comparison with X$threshold after the cross-distances are computed.
> However, I was wondering if there was another tool I should be
> considering.  Any and all thoughts are very welcome.  Thanks in  
> advance.
>
> Thanks,
> Allie
> -- 
> Alexander Shenkin
> PhD Candidate
> School of Natural Resources and Environment
> University of Florida
-- 
David Winsemius, MD
West Hartford, CT