nmds: Non-metric multidimensional scaling

In ecodist: Dissimilarity-Based Functions for Ecological Analysis

Non-metric multidimensional scaling

Description

Non-metric multidimensional scaling.

Usage

nmds(dmat, mindim = 1, maxdim = 2, nits = 10, iconf, epsilon = 1e-12,
    maxit = 500, trace = FALSE)

Arguments

dmat	lower-triangular dissimilarity matrix.
mindim	optional, minimum number of dimensions to use.
maxdim	optional, maximum number of dimensions to use.
nits	optional, number of separate ordinations to use.
iconf	optional, initial configuration. If not specified, then a random configuration is used.
epsilon	optional, acceptable difference in stress.
maxit	optional, maximum number of iterations.
trace	if TRUE, will write progress indicator to the screen.

Details

The goal of NMDS is to find a configuration in a given number of dimensions which preserves rank-order dissimilarities as closely as possible. The number of dimensions must be specified in advance. Because NMDS is prone to finding local minima, several random starts must be used. Stress is used as the measure of goodness of fit. A lower stress indicates a better match between dissimilarity and ordination. As of ecodist 1.9, the stress calculation used is the same as in MASS:isoMDS. In previous versions it was monotonically related, so the same configurations were produced, but the absolute value was different.

Value

conf	list of configurations, each in the same units as the original dissimilarities.
stress	list of final stress values.
r2	total variance explained by each configuration.

The first results are for the lowest number of dimensions (total number is (mindim - maxdim + 1) * nits).

Author(s)

Sarah Goslee

References

Kruskal, J.B. 1964. Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. Psychometrika 29:1-27.

Minchin, P.R. 1987. An evaluation of the relative robustness of techniques for ecological ordination. Vegetatio 96:89-108.

See Also

plot.nmds, min.nmds, vf, addord

Examples

data(iris)
iris.d <- dist(iris[,1:4])

### nmds() is timeconsuming, so this was generated
### in advance and saved.
### set.seed(1234)
### iris.nmds <- nmds(iris.d, nits=20, mindim=1, maxdim=4)
### save(iris.nmds, file="ecodist/data/iris.nmds.rda")
data(iris.nmds)

# examine fit by number of dimensions
plot(iris.nmds)

# choose the best two-dimensional solution to work with
iris.nmin <- min(iris.nmds, dims=2)

# rotate the configuration to maximize variance
iris.rot <- princomp(iris.nmin)$scores

# rotation preserves distance apart in ordination space
cor(dist(iris.nmin), dist(iris.rot))

# fit the data to the ordination as vectors
### vf() is timeconsuming, so this was generated
### in advance and saved.
### set.seed(1234)
### iris.vf <- vf(iris.nmin, iris[,1:4], nperm=1000)
### save(iris.vf, file="ecodist/data/iris.vf.rda")
data(iris.vf)

# repeat for the rotated ordination
### vf() is timeconsuming, so this was generated
### in advance and saved.
### set.seed(1234)
### iris.vfrot <- vf(iris.rot, iris[,1:4], nperm=1000)
### save(iris.vfrot, file="ecodist/data/iris.vfrot.rda")
data(iris.vfrot)

par(mfrow=c(1,2))
plot(iris.nmin, col=as.numeric(iris$Species), pch=as.numeric(iris$Species), main="NMDS")
plot(iris.vf)
plot(iris.rot, col=as.numeric(iris$Species), pch=as.numeric(iris$Species),
    main="Rotated NMDS")
plot(iris.vfrot)


# generate new data points to add to the ordination
# this might be new samples, or a second dataset

iris.new <- structure(list(Sepal.Length = c(4.6, 4.9, 5.4, 5.2, 6, 6.5, 6, 
6.8, 7.3), Sepal.Width = c(3.2, 3.5, 3.6, 2.3, 2.8, 3, 2.7, 3.1, 
3.2), Petal.Length = c(1.2, 1.5, 1.5, 3.5, 4.1, 4.2, 4.8, 5, 
5.7), Petal.Width = c(0.26, 0.26, 0.26, 1.2, 1.3, 1.4, 1.8, 2, 
2), Species = structure(c(1L, 1L, 1L, 2L, 2L, 2L, 3L, 3L, 3L), .Label = c("setosa", 
"versicolor", "virginica"), class = "factor")), .Names = c("Sepal.Length", 
"Sepal.Width", "Petal.Length", "Petal.Width", "Species"), class = "data.frame",
row.names = c(NA, -9L))

# provide a dist object containing original and new data
# provide a logical vector indicating which samples were used to
# construct the original configuration

iris.full <- rbind(iris, iris.new)
all.d <- dist(iris.full[,1:4])
is.orig <- c(rep(TRUE, nrow(iris)), rep(FALSE, nrow(iris.new)))

### addord() is timeconsuming, so this was generated
### in advance and saved.
### set.seed(1234)
### iris.fit <- addord(iris.nmin, iris.full[,1:4], all.d, is.orig, maxit=100)
### save(iris.fit, file="ecodist/data/iris.fit.rda")
data(iris.fit)

plot(iris.fit$conf, col=iris.full$Species, pch=c(18, 4)[is.orig + 1],
    xlab="NMDS 1", ylab="NMDS 2")
title("Demo: adding points to an ordination")
legend("bottomleft", c("Training set", "Added point"), pch=c(4, 18))
legend("topright", levels(iris$Species), fill=1:3)

ecodist documentation built on Nov. 2, 2023, 6:01 p.m.