Comparison of the methods of `hclust`

hclust_distplot.R

library(dplyr)
library(ggplot2)
x <- c("ward.D","ward.D2", "single","complete",
  "average","mcquitty","median","centroid")

X <- log(as.matrix(iris[,1:4]))

d <- dist(X)

dm <- as.matrix(d)

hres <- lapply(x, function(m)hclust(d,method=m))

do <- sapply(1:length(x),function(i)dm[hres[[i]]$order[1],hres[[i]]$order])

#s <- unname(dm[1,][order(dm[1,])])

colnames(do) <- x
row.names(do) <- NULL

df <- reshape2::melt(do, varnames=c("order","method")) %>% 
  mutate(method=factor(method,levels=x))

ggplot(df,aes(x=order,y=value))+
  geom_line()+
  facet_wrap(~method)+
  theme_bw()+labs(y="dist")
ggsave("dist.png")

###

df <- vector("list", length(x))
for(i in 1:length(x)){
  do <- dm[hres[[i]]$order,hres[[i]]$order]
  colnames(do) <- NULL
  rownames(do) <- NULL
  do[lower.tri(do,diag = TRUE)] <-NA 
  df[[i]] <- mutate(reshape2::melt(do,varnames = c("row","col")),
                    method=x[i]) %>% 
    dplyr::filter(!is.na(value))
}


df <- bind_rows(df) %>% 
  mutate(method=factor(method,levels=x))

ggplot(df,aes(x=col,y=row,fill=value))+
  geom_tile()+
  scale_fill_viridis_c()+
  facet_wrap(~method)+
  scale_y_reverse()+labs(fill="dist")

ggsave("distmat.png")