rng(x=NULL, dx=NULL, r = 1, method = NULL, usedeldir = TRUE, open = TRUE, k = NA, algorithm = 'cover_tree')
x
or dx
must be provided.dist
get.knn
is used from FNN to approximate
the relative neighborhood graph. Only the k
nearest neighbors to
the points are used to determine whether an edge should be made or not.
This will be much faster and use less memory for large data sets, but
is an approximation unless k
is sufficiently large.get.knn
.rng
.r
times this distance if r
is not 1). There is
an edge between the vertices associated with the observations if
and only if there are no vertices in the lune defined by the
intersection of the balls. The flag open
should make no difference for most applications,
but there are very specific cases (see the example section below)
where setting it to be TRUE will give the wrong answer (thanks to
Luke Mathieson for pointing this out to me).
J.W. Jaromczyk and G.T. Toussaint, "Relative neighborhood graphs and their relatives", Proceedings of the IEEE, 80, 1502-1517, 1992.
G.T. Toussaint, "A Graph-Theoretic Primal Sketch", Computational Morphology, 229-260, 1988.
D.J. Marchette, Random Graphs for Statistical Pattern Recognition, John Wiley & Sons, 2004.
gg
,cccd
,ccd
,
dist
get.knn
x <- matrix(runif(100),ncol=2) g <- rng(x) ## Not run: # plot(g) # ## End(Not run) ## Example using 'open': g <- graph.full(5,directed=FALSE) g1 <- rng(x=get.adjacency(g,sparse=FALSE),open=TRUE) ecount(g1) g2 <- rng(x=get.adjacency(g,sparse=FALSE),open=FALSE) graph.isomorphic(g2,g)