Raw <- read.csv("ptr_all_T.csv")
dim(Raw)
[1] 900  36
Scale_Raw_Mat <- matrix(scale(Raw[,2:36]))
#create som grid
Scale_Raw_Mat_Grid = somgrid(xdim = 30, ydim=30, topo="hexagonal")
#train som model
Scale_Raw_Mat_Som = som(Scale_Raw_Mat, grid=Scale_Raw_Mat_Grid, rlen=100, alpha=c(0.05,0.01))
#optomize the best number of clusters
#gererate distance matrix for map lacation
Dist_Map <- kohonen::unit.distances(Scale_Raw_Mat_Grid)
#gererate distance matrix for codes
Codes <- getCodes(Scale_Raw_Mat_Som)
Dist_Codes <- as.matrix(dist(Codes))
#exponentiate Euclidean distance by distance on map
Dist_Exp <- Dist_Codes^Dist_Map
#create plot to determine best number of clusters
factoextra::fviz_nbclust(Dist_Exp, factoextra::hcut, method =
"silhouette", hc_method = "ward.D2", k.max =  15)
Clust_Exp = hclust(as.dist(Dist_Exp), 'ward.D2')
Som_Cluster_Exp = cutree(Clust_Exp, 3)
#create color for som mapping

#plot som clusters om map
plot(Scale_Raw_Mat_Som, type="mapping", keep.data==TRUE, bgcol = col_vector[Som_Cluster_Exp], pchs = NA)`