vegan (version 2.7-1)

MDSaddpoints: Add New Points to NMDS ordination

Description

Add new points to an existing metaMDS or monoMDS ordination.

Usage

MDSaddpoints(nmds, dis, neighbours = 5, maxit = 200)
dist2xy(dist, pick, type = c("xy", "xx"), invert = FALSE)

Value

Function return a list of class "nmds" (there are no other objects of that type in vegan) with following elements

points

Coordinates of added new points

seeds

Starting coordinates for new points.

deltastress

Change of stress with added points.

iters

Number of iterations.

cause

Cause of termination of iterations. Integer for convergence criteria in monoMDS.

Arguments

nmds

Result object from metaMDS or monoMDS. The configuration of points is fixed, but new points are added.

dis

Rectangular non-symmetric dissimilarity matrix among new points (rows) and old fixed points (columns). Such matrix can be extracted from complete dissimilarities of both old and new points with dist2xy, or calculated with designdist2.

neighbours

Number of nearest points used to get the starting locations for new points.

maxit

Maximum number of iterations.

dist

Input dissimilarities.

pick

Indices (integers) of selected observations or a logical vector that is TRUE for picked items. The output will be in the original order and will not be reordered by this argument.

type

"xy" returns rectangular data of picked against not picked observations, and "xx" a subset of symmetric dissimilarities.

invert

Invert pick: drop elements listed.

Details

Function provides an interface to monoMDS Fortran code to add new points to an existing ordination that will be regarded as fixed. The function has a similar role as predict functions with newdata in Euclidean ordination (e.g. predict.cca). Input data must be a rectangular matrix of distances among new added points (rows) and all fixed old points (columns). Such matrices can be extracted from complete dissimilarities with helper function dist2xy. Function designdist2 can directly calculate such rectangular dissimilarity matrices between sampling units (rows) in two matries. In addition, analogue has distance function that can calculate dissimilarities among two matrices, including functions that cannot be specified in designdist2.

Great care is needed in preparing the dissimilarities for the input. The dissimilarity index must be exactly the same as in the fixed ordination, and columns must match old fixed points, and rows added new points.

Examples

## Cross-validation: remove a point when performing NMDS and add as
## a new points
data(dune)
d <- vegdist(dune)
## remove point 3 from ordination
mod3 <- metaMDS(dist2xy(d, 3, "xx", invert = TRUE), trace=0)
## add point 3 to the result
MDSaddpoints(mod3, dist2xy(d, 3))
## Use designdist2
d15 <- designdist(dune[1:15,])
m15 <- metaMDS(d15, trace=0)
MDSaddpoints(m15, designdist2(dune[1:15,], dune[16:20,]))