mapsize: Determining SOM size

In johnlpearce/sommix: Tools to Apply Self-Organizing Maps

Description

mapsize provides diagnostic plots and summaries of select criteria for determining how the size of the SOM influences characteristics of the SOM model.

Usage

mapsize(
  x,
  kmn = NULL,
  kmx = NULL,
  itermax. = NULL,
  nstarts. = NULL,
  maptopo. = NULL,
  distmet. = NULL,
  lmode. = NULL,
  symsize = 1
)

Arguments

x	is a data object to train the map. Should be a numerical or factor based data matrix.
kmn	is the minimum number of map units (aka nodes) to evaluate. Default value is 2.
kmx	is maximum number of map units (aka nodes) to evaluate. Default is 5*sqrt(n).
itermax.	maximum number of iterations passed to sommix. Default is 500*number of map nodes (k).
nstarts.	number of random initializatons passed to sommix. Default is 5.
maptopo.	map topology passed to sommix. Default is rectangular.
distmet.	distance method passed to sommix. Default is Euclidean.
lmode.	m initializatons passed to sommix
symsize	sets symbol size on plots

Details

An important step in the application of SOM are the user provided inputs for the dimensions of the mapping (i.e., size). Here we provide common model performance metrics and class-level evaluations in effort to assist the user in determining an appropriate map size.

Value

Panels a-d on the diagnostic plot illustrate common model perfomance metrics as a function of map size. Panels e-f examine within-class-sum-of-squares and frequency distributions as a function of map size. A list of model and class-level perfomance statistics is also returned.

• SOMX x dimension

• SOMY y dimension

• K Number of map nodes

• R2 R2

• ADJ_R2 Adjusted R2

• MAE mean absolute error based on class assignment distances

• RMSE root-mean-square-error based on class assignment distances

• AIC a form of Akaikes Information Criteria applied to clustering algorithms

• TotWCSS Total Within-Cluster Sum-of-Squares

• N Number class assignments

• FREQ Proportion of class assignments

• WCD average within-class distance

• BCD average between-class distacne

• WB_Ratio WCD/BCD

Examples

#NIEHS Mixtures Workshop dataset1
data(dataset1)
mapsize(scale(dataset1[,3:9]), kmx=10, itermax.=10)

johnlpearce/sommix documentation built on Jan. 7, 2021, 11:38 p.m.