Hi, I am using fanny and I have estrange results. I am wondering if someone out there can help me understand why this happens. First of all in most of my tries, it gives me a result in which each object has equal membership in all clusters. I have read that that means "the clustering is entirely fuzzy". Looking at the graphics it is really difficult to understand how objects with so different scores for the variables have the same membership for all the clusters. I also find estrange the fact that if I set K=3 (three clusters), it gives membership for all three clusters (0.333 for all of them) and then when it gives the closest hard clustering they only belong to cluster 1 or 2, but none of them to cluster three. The plot shows only two clusters (also the silhouette plot, even if it gives in the "silhouette plot information" the silhouette width for the three clusters????????. Then, for the same data I set k=4 and surprisingly, it gives membership for the four of them (this time they are not all the same) and when it gives the closest hard clustering they only belong to cluster 1, 2, or 3 but none of them to cluster 4. The plot shows only three clusters (also the silhouette plot, even if it gives in the "silhouette plot information" the silhouette width for the four clusters????????. why didn't it give this three clusters when I set k=3?????? For k=5 it gives all the information and then it only plots 2 clusters. This is very confusing. Also, if there is equal membership for all the clusters, how is it that I have a "closest hard clustering"? and a "neighbor"? Thank you in advance, Barbara
Barbara Diaz wrote:> Hi, > > I am using fanny and I have estrange results. I am wondering if > someone out there can help me understand why this happens. > > First of all in most of my tries, it gives me a result in which each > object has equal membership in all clusters. I have read that that > means "the clustering is entirely fuzzy". Looking at the graphics it > is really difficult to understand how objects with so different scores > for the variables have the same membership for all the clusters. > > I also find estrange the fact that if I set K=3 (three clusters), it > gives membership for all three clusters (0.333 for all of them) and > then when it gives the closest hard clustering they only belong to > cluster 1 or 2, but none of them to cluster three. The plot shows only > two clusters (also the silhouette plot, even if it gives in the > "silhouette plot information" the silhouette width for the three > clusters????????. > > Then, for the same data I set k=4 and surprisingly, it gives > membership for the four of them (this time they are not all the same) > and when it gives the closest hard clustering they only belong to > cluster 1, 2, or 3 but none of them to cluster 4. The plot shows only > three clusters (also the silhouette plot, even if it gives in the > "silhouette plot information" the silhouette width for the four > clusters????????. why didn't it give this three clusters when I set > k=3?????? > > For k=5 it gives all the information and then it only plots 2 clusters. > > This is very confusing. Also, if there is equal membership for all the > clusters, how is it that I have a "closest hard clustering"? and a > "neighbor"?Can you speicfy a reproducible examples, please? Without an example, it is really hard to explain what happens... Uwe Ligges> Thank you in advance, > > Barbara > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
> Barbara Diaz wrote: > > Hi, > > > > I am using fanny and I have estrange results. I am wondering if > > someone out there can help me understand why this happens. > > > > First of all in most of my tries, it gives me a result in > which each > > object has equal membership in all clusters. I have read that that > > means "the clustering is entirely fuzzy". Looking at the > graphics it > > is really difficult to understand how objects with so > different scores > > for the variables have the same membership for all the clusters.Hi Barbara, I think, there is a problem with fanny, when you have standardised data. For example: library(mvoutlier) library(cluster) data(chorizon) a <- fanny(chorizon[,101:110],4) b <- fanny(scale(chorizon[,101:110]),4) a$mem # is ok, but b$mem # have same memberships Better to use function cmeans in package e1071, which gives correct memberships! Best, Matthias> ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read > the posting guide! http://www.R-project.org/posting-guide.html >