Evaluating HCPC clusters from FactoMineR using cluster.stats from fpc library
0
0
Entering edit mode
@anahochmanova-21482
Last seen 4.7 years ago

I'm trying to get quality measures, for example silhouette, from HCPC clusters

I can get avg.silwidth from the k-means algorithm, using cluster.stats from fpc library :

library(factoextra)
library(fpc)
km.res <- eclust(dat, "kmeans", k = 20, nstart = 25, graph = FALSE)
# Compute pairwise-distance matrices
dd <- dist(dat, method ="euclidean")
# Statistics for k-means clustering
km_stats <- cluster.stats(dd,  km.res$cluster)

But now I'm trying to use cluster.stats with HCPC. To run cluster.stats I need a distance matrix and the clusters. When I execute HCPC with consol=TRUE it indeed runs a k-mean algorithm therefore I've been trying to find the distance matrix and the clusters but I was not able.

names(res.hcpc) [1] "data.clust" "desc.var" "desc.axes" "call" "desc.ind"

In this assignment https://eric.univ-lyon2.fr/~jahpine/cours/l3_cestat-dm/tp5.pdf

it seems that they get the clusters from data.clust$clust:

clust . eval . hcpc = cluster . stats ( cluster :: daisy ( sub . D . kmodes ) , as . numeric ( res . hcpc . sub . acm$data . clust$clust ) , res . kmodes . sub . D . kmodes$cluster )

But when I run this

library(FactoMineR)
res <- PCA(dat,quali.sup=1:3, ncp=8)
#Hierarchical ascendant clustering
res.hcpc <- HCPC(res, kk=Inf, min=3, max=10, consol=TRUE)
km_stats <- cluster.stats(dd, res.hcpc$data.clust$clust)

I get the Error:

Error in Summary.factor(c(3L, 1L, 1L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L,  : 
  ‘max’ not meaningful for factors
clustering r fpc FactoMineR • 1.9k views
ADD COMMENT
0
Entering edit mode

Hi,

I am just asking myself if you could find a solution for this? I am also dealing with the same issue...I want to get some cluster validation statistics for my HCPC but I just can't find any solution.

BR, Daniela

ADD REPLY

Login before adding your answer.

Traffic: 972 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6