Solved – Visualizing kmodes clusters

categorical dataclusteringdata visualization

I am working on cluster analysis of a completely categorical data set using package klaR and function kmodes. A sample of the data is available on dropbox. Just cross the sign-up notification dropbox will show when link opens.

The code to do the clustering was simple enough.

require(klaR)    
c1 <- kmodes(df, 5, 5, weighted = FALSE)

My questions are:

How do I visualize these clusters? I have done simple plots in the past with k-means clusters: plotcluster(data, clus$cluster). When I try this here, I get:
```
 Error: is.numeric(x) || is.logical(x) is not TRUE
```
How do I decide optimal number of clusters? I've read through Cluster analysis in R: determine the optimal number of clusters on Stack Overflow, but there is no mention of categorical variables anywhere and I could not understand which of the several methods discussed by the author will be applicable in my case.

Best Answer

I think you need the command plot(data, col=clus\$cluster) instead. Or rather just plot(data[,c(j,l)], col=clus$cluster). This will give the graphs of a group of columns with respect to the clusters. About optimal number of clusters, I would just try different number of clusters starting from 2 and try to see from there how good I can do.

Related Solutions

Solved – Visualizing large dataset with multiple subgroups

Seems like kind of a tall order, but here's a whirlwind tour of R.

library(party)
library(rattle)
library(ggplot2)
library(car)

#this will expand your test set so that it is large enough to generate a tree.
DO <- rbind(DO, DO, DO, DO)
DO.ctree <- ctree(DO ~ ., data = DO, 
               controls = ctree_control(maxsurrogate = 3))
plot(DO.ctree)
#I think this answers both your "first and foremost" and your "main" questions.
#In brief: The party package helps identify which variables most influence the 
#dependent variable

ctree output

ggplot(DO, aes(factor(Season), DO)) + geom_point()
#lots of easy descriptive stats in ggplot package

dotplot from ggplot2

DO <- DO[, !sapply(DO, function (x) all(is.na(x)))]
DO.numeric <- DO[ ,sapply(DO, is.numeric)]
round(cor(na.omit(DO.numeric)), 1)
#           T   SC   pH DO.S   DO  Tur Flow Rainfall Solar
# T         1.0  1.0  1.0  0.7  0.3 -0.8 -0.9     -1.0   0.0
# SC        1.0  1.0  1.0  0.7  0.3 -0.9 -0.9     -1.0   0.1
# pH        1.0  1.0  1.0  0.7  0.3 -0.8 -0.8     -1.0   0.0
# DO.S      0.7  0.7  0.7  1.0  0.9 -0.9 -0.9     -0.6   0.7
# DO        0.3  0.3  0.3  0.9  1.0 -0.7 -0.6     -0.1   0.9
# Tur      -0.8 -0.9 -0.8 -0.9 -0.7  1.0  1.0      0.8  -0.6
# Flow     -0.9 -0.9 -0.8 -0.9 -0.6  1.0  1.0      0.8  -0.5
# Rainfall -1.0 -1.0 -1.0 -0.6 -0.1  0.8  0.8      1.0   0.1
# Solar     0.0  0.1  0.0  0.7  0.9 -0.6 -0.5      0.1   1.0
#Here's a brief corelation summary

scatterplotMatrix(na.omit(DO.numeric))
#Here's the big chart of correlations I think you requested

scatterplotMatrix

You may be interested in checking out the rattle package/GUI: it can get you off to a quick start with a lot these general questions.

Solved – How to measure cluster quality with distance matrix

If you read the section on internal cluster validation in Wikipedia, you will learn about a dozen measures for evaluation that require paiwise distances. E.g.

Silhouette index
Dunn's index
Davids-Bouldin

and many more.

On the other hand, none of them has me really convinced yet.

Best Answer

Related Solutions

Solved – Visualizing large dataset with multiple subgroups

Solved – How to measure cluster quality with distance matrix

Related Question