nearestdist.Rd
This function computes and returns specific elements of distance matrix computed by using the specified distance measure.
nearest.dist( x, y=NULL, method = "euclidean",
delta = 1, upper = if (is.null(y)) FALSE else NULL,
p = 2, miles = TRUE, R = NULL, fortran = FALSE)
Matrix of first set of locations where each row gives the coordinates of a particular point. See also ‘Details’.
Matrix of second set of locations where each row gives the
coordinates of a particular point. If this is missing x
is
used. See also ‘Details’.
the distance measure to be used. This must be one of
"euclidean"
, "maximum"
, "minkowski"
or
"greatcircle"
. Any unambiguous substring can be
given.
only distances smaller than delta
are recorded,
see Details.
Should the entire matrix (NULL
) or only the upper-triagonal (TRUE
)
or lower-triagonal (FALSE
) values be calculated.
The power of the Minkowski distance.
For great circle distance: If true distances are in statute miles if false distances in kilometers.
For great circle distance: Radius to use for sphere to find spherical distances. If NULL
the radius is either in miles or kilometers depending on the
values of the miles argument. If R=1
then distances are of
course in radians.
Should the C++ (FALSE
) or the Fortran code (TRUE
) be used. If 64-bit matrices are needed, the argument is set to (TRUE
).
A spam
object containing the distances spanned between
zero and delta
. The sparse matrix may contain many zeros
(e.g., collocated data). However, to calculate covariances, these zeros
are essential.
For great circle distance, the matrices x
and y
contain the degrees longitudes in the first and the degrees latitudes
in the second column. delta
is in
degrees. Hence to restrict to distances smaller than delta.km
,
one has to specify delta=delta.km*360/(6378.388*2*pi)
.
The distances are calculated based on spherical law of cosines.
Care is needed for `zero' distances due to the final acosin:
acos(1-1e-16)
, especially with an actual radius.
Default value of Earth's radius is 3963.34miles (6378.388km).
There are many other packages providing distance functions. Especially
for great circle distances there are considerable differences between
the implementations. For high precision results, sp::spDists
is
a good candidate and distances of large amount of locations can be
processed in parallel with the parallelDist
package.
The formerly depreciated arguments eps
and diag
are now
eliminated.
x
and y
can be any object with an existing
as.matrix
method.
The Fortran code is based on a idea of Doug Nychka.
# Note that upper=T and using t(X)+X is quicker than upper=NULL;
# upper=T marginally slower than upper=F.
# To compare nearest.dist with dist, use as.dist(...)
nx <- 4
x <- expand.grid(as.double(1:nx),as.double(1:nx))
sum( ( as.dist(nearest.dist( x, delta=nx*2))-
dist(x) )^2)
#> [1] 0
# Create nearest neighbor structures:
par(mfcol=c(1,2))
x <- expand.grid(1:nx,1:(2*nx))
display( nearest.dist( x, delta=1))
#> Warning: default value for 'cex' in 'display' might not be the optimal choice
x <- expand.grid(1:(2*nx),1:nx)
display( nearest.dist( x, delta=1))
#> Warning: default value for 'cex' in 'display' might not be the optimal choice