Re: [SOCIAL]Question about Ward hierarchical clustering in function HCPC

[Francois Husson]

Hi,

In the program, before using hclust, I first calculate the square distance, that is why I use method="ward" with hclust.
It is thus equivalent to use "ward.D2" on the distance.

FH

Le 08/06/2023 à 17:53, Jhan Rodriguez a écrit :

Hi,

I have seen in the code that the HCPC function uses flashClust::hclust for hierarchical clustering, with method=”ward”.

This seems to be the same as using stats::hclust() with method “ward.D”:

myDist <- dist(USArrests)

plot(stats::hclust(myDist, method = "ward.D"))plot(flashClust::hclust(myDist, method = "ward"))

 

In view of the documentation for function hclust{stats} quoted below, wouldn’t it be rather better to use stats::hclust(myDist, method = "ward.D2") for the hierarchic clustering part?

 

I mean this part of the documentation:

„

Two different algorithms are found in the literature for Ward clustering. The one used by option "ward.D" (equivalent to the only Ward option "ward" in R versions \le≤ 3.0.3) does not implement Ward's (1963) clustering criterion, whereas option "ward.D2" implements that criterion (Murtagh and Legendre 2014). With the latter, the dissimilarities are squared before cluster updating. Note that agnes(*, method="ward") corresponds to hclust(*, "ward.D2").

“

Thanks a lot.

Regards

JR

--
Vous recevez ce message, car vous êtes abonné au groupe Google Groupes "FactoMineR users".
Pour vous désabonner de ce groupe et ne plus recevoir d'e-mails le concernant, envoyez un e-mail à l'adresse factominer-use...@googlegroups.com.
Cette discussion peut être lue sur le Web à l'adresse https://groups.google.com/d/msgid/factominer-users/9553ea6b-4a0e-44f6-9e22-38409ae60198n%40googlegroups.com.

--

	François Husson Department Statistics & Computer science UMR 6625 IRMAR CNRS 65 rue de Saint-Brieuc, CS 84215, 35042 Rennes Cedex Tel: +33 2 23 48 58 86 https://husson.github.io
En 2022, Agrocampus Ouest devient l'Institut Agro Rennes-Angers.