vegan (version 1.0-1)

initMDS: Random Start for isoMDS

Description

Random start for multidimensional scaling.

Usage

initMDS(x, k=2)

Arguments

x

Dissimilarity matrix for isoMDS.

k

Number of dimensions.

Value

• A random configuration which is intended to be used within isoMDS only.

Details

Returns a random configuration for multidimensional scaling. The default practice in isoMDS is to start with metric scaling, but this is easily trapped in a local minimum. Thus the recommended way is to use several random starts and select a solution with lowest stress and similar configuration to other solutions with about similar stress.

References

Minchin, P.R. (1987) An evaluation of relative robustness of techniques for ecological ordinations. Vegetatio 71, 145-156.

See Also

isoMDS, cmdscale, procrustes.

Examples

library(mva)
library(MASS)
data(varespec)
vare.dist <- vegdist(varespec, "bray")
mds.null <- isoMDS(vare.dist)
## See if you can get any better
## (This is a small data set, and an easy case: probably you can't
## improve much).
stress <- rep(NA,10)
for (i in 1:10)
  stress[i] <- isoMDS(vare.dist, initMDS(vare.dist), maxit=200, trace=F)