Colorize Clusters in Dendogram with ggplot2

问题:

Didzis Elferts showed how to plot a dendogram using ggplot2 and ggdendro:

horizontal dendrogram in R with labels

here is the code:

labs = paste("sta_",1:50,sep="") #new labels
rownames(USArrests)<-labs #set new row names
hc <- hclust(dist(USArrests), "ave")

library(ggplot2)
library(ggdendro)

#convert cluster object to use with ggplot
dendr <- dendro_data(hc, type="rectangle") 

#your own labels are supplied in geom_text() and label=label
ggplot() + 
  geom_segment(data=segment(dendr), aes(x=x, y=y, xend=xend, yend=yend)) + 
  geom_text(data=label(dendr), aes(x=x, y=y, label=label, hjust=0), size=3) +
  coord_flip() + scale_y_reverse(expand=c(0.2, 0)) + 
  theme(axis.line.y=element_blank(),
        axis.ticks.y=element_blank(),
        axis.text.y=element_blank(),
        axis.title.y=element_blank(),
        panel.background=element_rect(fill="white"),
        panel.grid=element_blank())

Does anyone know, how to colorize the different clusters? For example, you want to have 2 Clusters (k=2) colorized?

回答1:

Workaround would be to plot cluster object with plot() and then use function rect.hclust() to draw borders around the clusters (nunber of clusters is set with argument k=). If result of rect.hclust() is saved as object it will make list of observation where each list element contains observations belonging to each cluster.

plot(hc)
gg<-rect.hclust(hc,k=2)

Now this list can be converted to dataframe where column clust contains names for clusters (in this example two groups) - names are repeated according to lengths of list elemets.

clust.gr<-data.frame(num=unlist(gg),
  clust=rep(c("Clust1","Clust2"),times=sapply(gg,length)))
head(clust.gr)
      num  clust
sta_1   1 Clust1
sta_2   2 Clust1
sta_3   3 Clust1
sta_5   5 Clust1
sta_8   8 Clust1
sta_9   9 Clust1

New data frame is merged with label() information of dendr object (dendro_data() result).

text.df<-merge(label(dendr),clust.gr,by.x="label",by.y="row.names")
head(text.df)
   label  x y num  clust
1  sta_1  8 0   1 Clust1
2 sta_10 28 0  10 Clust2
3 sta_11 41 0  11 Clust2
4 sta_12 31 0  12 Clust2
5 sta_13 10 0  13 Clust1
6 sta_14 37 0  14 Clust2

When plotting dendrogram use text.df to add labels with geom_text() and use column clust for colors.

ggplot() + 
  geom_segment(data=segment(dendr), aes(x=x, y=y, xend=xend, yend=yend)) + 
  geom_text(data=text.df, aes(x=x, y=y, label=label, hjust=0,color=clust), size=3) +
  coord_flip() + scale_y_reverse(expand=c(0.2, 0)) + 
  theme(axis.line.y=element_blank(),
        axis.ticks.y=element_blank(),
        axis.text.y=element_blank(),
        axis.title.y=element_blank(),
        panel.background=element_rect(fill="white"),
        panel.grid=element_blank())

回答2:

This approach is very similar to @DidzisElferts', just a little simpler.

df   <- USArrests                 # really bad idea to muck up internal datasets
labs <- paste("sta_",1:50,sep="") # new labels
rownames(df) <- labs              # set new row names

library(ggplot2)
library(ggdendro)
hc       <- hclust(dist(df), "ave")           # heirarchal clustering
dendr    <- dendro_data(hc, type="rectangle") # convert for ggplot
clust    <- cutree(hc,k=2)                    # find 2 clusters
clust.df <- data.frame(label=names(clust), cluster=factor(clust))
# dendr[["labels"]] has the labels, merge with clust.df based on label column
dendr[["labels"]] <- merge(dendr[["labels"]],clust.df, by="label")
# plot the dendrogram; note use of color=cluster in geom_text(...)
ggplot() + 
  geom_segment(data=segment(dendr), aes(x=x, y=y, xend=xend, yend=yend)) + 
  geom_text(data=label(dendr), aes(x, y, label=label, hjust=0, color=cluster), 
           size=3) +
  coord_flip() + scale_y_reverse(expand=c(0.2, 0)) + 
  theme(axis.line.y=element_blank(),
        axis.ticks.y=element_blank(),
        axis.text.y=element_blank(),
        axis.title.y=element_blank(),
        panel.background=element_rect(fill="white"),
        panel.grid=element_blank())

回答3:

Adding to @DidzisElferts' and @jlhoward's code, the dendrogram itself can be coloured.

library(ggplot2)
library(ggdendro)
library(plyr)
library(zoo)

df <- USArrests                       # really bad idea to muck up internal datasets
labs <- paste("sta_", 1:50, sep = "") # new labels
rownames(df) <- labs                  # set new row names

cut <- 4    # Number of clusters
hc <- hclust(dist(df), "ave")              # hierarchical clustering
dendr <- dendro_data(hc, type = "rectangle") 
clust <- cutree(hc, k = cut)               # find 'cut' clusters
clust.df <- data.frame(label = names(clust), cluster = clust)

# Split dendrogram into upper grey section and lower coloured section
height <- unique(dendr$segments$y)[order(unique(dendr$segments$y), decreasing = TRUE)]
cut.height <- mean(c(height[cut], height[cut-1]))
dendr$segments$line <- ifelse(dendr$segments$y == dendr$segments$yend &
   dendr$segments$y > cut.height, 1, 2)
dendr$segments$line <- ifelse(dendr$segments$yend  > cut.height, 1, dendr$segments$line)

# Number the clusters
dendr$segments$cluster <- c(-1, diff(dendr$segments$line))
change <- which(dendr$segments$cluster == 1)
for (i in 1:cut) dendr$segments$cluster[change[i]] = i + 1
dendr$segments$cluster <-  ifelse(dendr$segments$line == 1, 1, 
             ifelse(dendr$segments$cluster == 0, NA, dendr$segments$cluster))
dendr$segments$cluster <- na.locf(dendr$segments$cluster) 

# Consistent numbering between segment$cluster and label$cluster
clust.df$label <- factor(clust.df$label, levels = levels(dendr$labels$label))
clust.df <- arrange(clust.df, label)
clust.df$cluster <- factor((clust.df$cluster), levels = unique(clust.df$cluster), labels = (1:cut) + 1)
dendr[["labels"]] <- merge(dendr[["labels"]], clust.df, by = "label")

# Positions for cluster labels
n.rle <- rle(dendr$segments$cluster)
N <- cumsum(n.rle$lengths)
N <- N[seq(1, length(N), 2)] + 1
N.df <- dendr$segments[N, ]
N.df$cluster <- N.df$cluster - 1

# Plot the dendrogram
ggplot() + 
   geom_segment(data = segment(dendr), 
      aes(x=x, y=y, xend=xend, yend=yend, size=factor(line), colour=factor(cluster)), 
      lineend = "square", show.legend = FALSE) + 
   scale_colour_manual(values = c("grey60", rainbow(cut))) +
   scale_size_manual(values = c(.1, 1)) +
   geom_text(data = N.df, aes(x = x, y = y, label = factor(cluster),  colour = factor(cluster + 1)), 
      hjust = 1.5, show.legend = FALSE) +
   geom_text(data = label(dendr), aes(x, y, label = label, colour = factor(cluster)), 
       hjust = -0.2, size = 3, show.legend = FALSE) +
   scale_y_reverse(expand = c(0.2, 0)) + 
   labs(x = NULL, y = NULL) +
   coord_flip() +
    theme(axis.line.y = element_blank(),
        axis.ticks.y = element_blank(),
        axis.text.y = element_blank(),
        axis.title.y = element_blank(),
        panel.background = element_rect(fill = "white"),
        panel.grid = element_blank())

The 2-cluster and 4-cluster solutions:

回答4:

A short way to achieve a similar result is to use the package dendextend (derived from this nice overview).

df   <- USArrests   # really bad idea to muck up internal datasets
labs <- paste("sta_",1:50,sep="") # new labels
rownames(df) <- labs # set new row names

require(magittr)
require(ggplot)
require(dendextend)

dend <- df %>% dist %>%
  hclust %>% as.dendrogram %>%
  set("branches_k_color", k = 4) %>% set("branches_lwd", 0.7) %>%
  set("labels_cex", 0.6) %>% set("labels_colors", k = 4) %>%
  set("leaves_pch", 19) %>% set("leaves_cex", 0.5) 
ggd1 <- as.ggdend(dend)
ggplot(ggd1, horiz = TRUE)

Note: The order of the states is slightly different compared to those above - not really changing interpretation though.