boostMDS: Improve goodness-of-fit of a given MDS solution in terms of...

Description Usage Arguments Value References Examples

View source: R/mds.R

Description

Given a distance matrix and a valid MDS representation for it, improve the R-square correlation between observed and approximated distances until converged is reached for a given threshold.

Usage

1
2
boostMDS(D, Y, rate = 0.01, maxit = 50, tol = 0.001, samplesize,
verbose = TRUE, scale = FALSE, seed = 149, plt = FALSE, mc.cores = 1)

Arguments

D

Distance matrix.

Y

Matrix with points from a valid MDS solution for the distances in D.

rate

Grid step rate, start with 0.1 which usually is a good compromise, try also 0.01, 1, 10.

maxit

Maximum number of iterations.

tol

Tolerace for R-square convergence.

samplesize

When there are over 100 points to represent, the gradiend descent step size is determined using a fraction samplesize of the original data points. By default 0.01 with a minimum of 100 points, which typically gives very stable results. Setting large samplesize can significantly increase the computational cost.

verbose

Give details of the gains in R-square and step size.

scale

Whether to scale the MDS coordinates in the output MDS.

seed

A random seed to be used in the resampling process if samplesize < 1.

plt

Whether to plot the intermediate solutions or not.

mc.cores

Number of cores to use in parallelized grid step size search.

Value

The function returns a matrix with the coordinates of a valid MDS solution for distance matrix D where the R-square correlation has been improved. However, have in mind that an MDS solution with better R-square does not necessarily mean the solution is easier to interpret. As with any MDS approach, a balance must be found between pure 'technical' goodness-of-fit and usefulness of the delivered solution in terms of answering the original hypothesis.

References

boostMDS is based on hitMDS (High-Throughput Multidimensional Scaling, see see http://dig.ipk-gatersleben.de/hitmds/hitmds.html for details)

Examples

1
2
3
4
5
6
7
8
# Not run, see also chroGPS-manual.pdf file for examples
#data(geneSample)
#d = distGPS(geneSample,uniqueRows=TRUE)
#m = mds(d,type='isoMDS')
#m
#plot(m)
#m = boostMDS([email protected],[email protected])
#plot(m)

Bioconductor-mirror/chroGPS documentation built on June 1, 2017, 5:32 a.m.