如何获取每个节点中 SOM 图中的簇数？

Question

I want to understand to which node my wine is connected after getting a som plot.

That's why firstly we need to get data.frame with the name of wine and the number of cluster that wines belongs to. And next step would be to see the number of the cluster on this plot. But idk how:)

data(wines)
View(wines)    
#adding id for each wine

wines<-as.data.frame(wines)
wines$ID <- seq.int(nrow(wines))

#substract the id to know the "name" of wine

som_wines<-wines[,-14]
som_model<-som(scale(som_wines), grid = somgrid(5, 5, "hexagonal"))
som_codes<-as.data.frame(som_model$codes)

#ilustrating needed quantity of clusters

mydata <- as.data.frame(som_model$codes)
wss <- (nrow(mydata)-1)*sum(apply(mydata,2,var)) 
for (i in 2:15) {
  wss[i] <- sum(kmeans(mydata, centers=i)$withinss)
}
plot(wss)

#som plot

som_cluster <- cutree(hclust(dist(som_codes)), 3)
plot(som_model, type="codes",bgcol= som_cluster, main = "Clusters") 
add.cluster.boundaries(som_model, som_cluster)   ` 

#Here we got 3 clusters. Creating the dataframe which defines wines id's to cluster groups.

cluster_details <- data.frame(id=wines$ID, cluster=som_cluster[som_model$unit.classif])

And now I want numbers of clusters to be shown there, on the som plot. Are there any suggestions how to cope with that? Would appreciate any answer :)

Answer 1

You can check to what node each observation belongs to by calling the model's variable namely unit.classif. Based on your scripts, you assign the model to som_model. Therefore, you can call

som_model$unit.classif

The vector is ordered by the row order of your data, ie your 1st input data belongs to the node denoted by the 1st unit.classif vector value, and so on. You can check by calling

length(som_model$unit.classif)
nrow(som_wines)

They have the same length. The library arrange the nodes in a matrix whose number of dimension of (nodes x features). If you define your model to have 5x5 nodes, whilst your data have 13 features, so your model's node would be denoted as a 25x13 matrix. You chan check by calling

dim(som_model$codes[[1]])

On the map, the nodes line up from bottom left to top right. The first node is on the bottom left and the 25th node is on the top right of the codes map. So if you want to know the position of the node to which a particular data belongs to, you can extend your script to something like this:

from.bottom <- ceiling(som_model$unit.classif / som_model$grid$xdim)
from.left <- som_model$unit.classif %% som_model$grid$xdim
from.left[from.left == 0] <-  som_model$grid$xdim

cluster_details <- cbind(
  cluster_details, som.unit = som_model$unit.classif,
  from.bottom = from.bottom, from.left = from.left
)

(cluster_details)

Answer 2

the answer is situated here: add clusters and nodes from SOMbrero package to training data

Particularly in these lines :

SomModel <- som(
    data = TrainingMatrix,
    grid = GridDefinition,
    rlen = 10000,
    alpha = c(0.05, 0.01),
    keep.data = TRUE
)

nb <- table(SomModel$unit.classif)
groups = 5
tree.hc = cutree(hclust(d=dist(SomModel$codes[[1]]),method="ward.D2",members=nb),groups)


result <- OrginalData
result$Cluster <- tree.hc[SomModel$unit.classif]
result$X <- SomModel$grid$pts[SomModel$unit.classif,"x"]
result$Y <- SomModel$grid$pts[SomModel$unit.classif,"y"]