sommix_eval: Generate multiple plots of SOM model results
In johnlpearce/sommix: Tools to Apply Self-Organizing Maps

Description Usage Arguments Details Value

Plots SOM output using Parallel Coordinate Plots, Multi-dimensional Scaling (MDS), and Sammons mapping. kmeans, wards, and PCA

1	sommix_eval(som_obj, labtype = "IDs", labsize = 1, legcex = 1, colormod = NULL)

`som_obj`	a SOM object
`labtype`	determines label type. "IDs" and "XYs" available.
`labsize`	sets label size
`legcex`	sets legend cex
`colormod`	sets colorscheme

Additional visualization of SOM results can lead to stronger interpretion. Panel (a) presenst a parallel coordinates plot of SOM profile results in order visualize differences in the individual components. Each vertical bar represents a variable axis where minimum values are on the bottom and maximum values are on top. A line is then used to connect component values for each profile. Panel (b) presents a multidimensional scaling plot that allows visualization of how well the SOM profile results span the original data space and how they compare to profiles generated from kmeans and Ward's heirarchical clustering. Panel (c) presents a Sammon's mapping of SOM profiles that effectively visualizes the the SOM using a distance preserving map projection. This provides a clearer picture of similarity/dissimilarity among SOM profiles.

Panel a-c plot SOM profile results using a range of techniques designed to enhance understanding of results. A list is also returned that includes the following items.

EVALTAB a dataframe containing classification assignments for SOM, kmeans (KM), Ward (HC). Principal component scores for PC1 and PC2 are also included.
SOM_Profiles SOM profiles (a.k.a., codebook vectors)
KM_Profiles K-means profiles (a.k.a., cluster centers)
HC_Profiles Ward's hierarchical clustering profiles (a.k.a., cluster centers)
PC1_Loadings loadings from the first principal component
PC2_Loadings loadings from the second principal component
CLASSIC_AGREEMENT measures of class agreement ranging from 0:1 where 1 reflects perfect agreement