R/orderingKM.R

In vegan: Community Ecology Package

`orderingKM` <-
    function(mat)
{

### INPUT :
### mat		(n x k): n the objects and k the descriptors
### 			This matrix must be integers and numeric
###			And must not be binairy, it is the partition matrix
###			output by cascadeKM
### OUTPUT :  Ordered matrix

    ## Uses alternatively the fast USEPOWERALGORITHM provided by Legendre
    ## et al., or the standard R function cmdscale with matching
    ## coefficient provided by vegdist.c as method=50.

    USEPOWERALGORITHM <- TRUE

    ##Check up

    if(!is.matrix(mat)) stop("'mat' must be a matrix!")
    if(!is.numeric(mat)) stop("'mat' must be numeric!")
    if(any(is.na(mat))) stop("'NA' value was found in the matrix!")
    if(any(is.infinite(mat))) stop("'Inf' value was found in the matrix!")
    nb.desc=ncol(mat)
    nb.obj=nrow(mat)

    scores<-rep(0.0,nb.obj)
    if (USEPOWERALGORITHM) {
	scores<-as.vector(.Fortran("orderdata",as.integer(mat),
                                   as.integer(nb.obj), as.integer(nb.desc),
                                   sc=as.double(scores),
                                   PACKAGE="vegan")$sc)
    } else {
        d <- .C("veg_distance", x = as.double(mat), nr = nb.obj,
                nc = nb.desc, d = double(nb.obj * (nb.obj - 1)/2),
                diag = as.integer(FALSE), method = as.integer(50),
                PACKAGE = "vegan")$d
        attr(d, "Size") <- nb.obj
        attr(d, "Labels") <- dimnames(mat)[[1]]
        attr(d, "Diag") <- FALSE
        attr(d, "Upper") <- FALSE
        attr(d, "method") <- "matching"
        class(d) <- "dist"
        scores <- cmdscale(d, k = 1)[,1]

    }
    scores <- order(scores)
    mat<-mat[scores,]
    mat
}