[R-sig-Geo] Calculate shortest distance between points belonging to different polygons

[Ivan Palmegiani]

Dear members of the list,

I'm handling a SpatialPointsDataFrame with 100 ramdom points distributed 
within 13 different polygons.

 > randp
                     coordinates   Point_ID   Polygon_ ID
0  (690926.8, 7522595)         1_hs                    13
1  (696727.1, 7576122)         2_hs                      6
...
...
98 (728199.9, 7549810)     99_hs                    12
99 (723428.1, 7545891)   100_hs                    12

I need to calculate the shortest distance between points belonging to 
different polygons. Basically I'd like to do what nndist {spatstat} 
does. The difference is that the distance should be calculated between 
groups of points instead of within a group of points.

I tried to use "aggregate" as suggested below but it didn't work out for me.
http://www.inside-r.org/packages/cran/spatstat/docs/nndist

Please find my try below:

 > randp.df<-data.frame(randp)
 > randp.hs.df
        Point_ID     coords.x1      coords.x2   Polygon_ ID
0            1_hs     690926.8       7522595 13
1            2_hs     696727.1       7576122 6
2            3_hs     723480.7       7546594 12

library(spatstat)

# Calculate nearest neighbors within a polygon
 > nn.within.pol<-nndist(randp.df[,c(2,3)],by=marks(randp.df$Polygon_ID))
 > nn.within.pol
  [1]  2579.42199  1391.88915    59.85628   734.95108   734.95108 
840.65125   957.47838   741.58160   955.26483  3307.59444 1361.64626  
2682.70690
  ...
  ...
  [97]  1349.88694   955.26483  3166.00894   705.25663
# Ok but these are not the distances I need

# Calculate nearest neighbors between polygons
nn.between.pol<-aggregate(nn.within.pol, 
by=list(from=marks(randp.df$Polygon_ID)), min)
# Error in aggregate.data.frame(as.data.frame(x), ...) : arguments must 
have same length

 > nn.between.hs<-aggregate(randp.hs.df[,c(2,3)], 
by=list(randp.df$Polygon_ID), nndist)
 > nn.between.hs

The outcome is an asymmetric data frame (dim 13, 6) with a lot of empty 
cells and values that look unlikely to be distances.

The result I'd like to get is a matrix (dim 100, 1) with the distances 
between each random point and its nearest neighbor belonging to a 
different polygon (i.e. its nearest neighbor having a different Polygon_ID).

Can someone kindly correct my script or suggest a function able to do 
the job?

Cheers,

Ivan