Minstress: Better Starting Configuration For Non-Metric MDS

In analytics: Regression Outlier Detection, Stationary Bootstrap, Testing Weak Stationarity, NA Imputation, and Other Tools for Data Analysis

Description

Minstress is a heuristic to find better non-metric MDS solutions, by finding better starting configurations, instead of just using a random one.

Usage

Minstress(x, p, s, k, iter = 5, pb = F, m = "euclidean")

Arguments

x	a data frame containing numeric values only
p	the size of the population of seeds (any positive integer)
s	the number of seeds we sample (any positive integer)
k	the number of dimensions wanted (any positive integer)
iter	a positive integer specifying the number of iterations.
pb	a Boolean variable declaring if one wants to display a pogress bar (default: False)
m	a string specifying the distance method (default: 'euclidean')

Details

This function performs several iterations, each using a different starting seed, and in turn each one of those iterations performs non-metric MDS many times (typically, thousands or more) in an attempt to find the best seed (which induces a particular initial configuration) of them all.

Value

A list informing about dimensionality, minimum STRESS level found, and best seed found. One can then use the best seed found to perform non-metric MDS with a better initial configuration (generally).

Author(s)

Albert Dorador

Examples

require(MASS)

swiss.x <- as.data.frame(swiss[, -1])
Minstress(swiss.x, 1e5, 50, 2, iter = 3)

# Comparing without using Minstress (for such a low value of s, difference is minimal)
swiss.x <- as.matrix(swiss[, -1])
swiss.dist <- dist(swiss.x)
swiss.mds <- isoMDS(swiss.dist)

analytics documentation built on May 2, 2019, 3:37 p.m.