R：聚类树状图的ggplot高度调整
R: ggplot height adjustment for clustering dendrogram
想法是结合 R 包 ClustOfVar 和 ggdendro 给出变量聚类的可视化摘要。
当数据列数较少时，除了有未覆盖的区域（如下图圈出）外，效果还不错。使用 mtcars 例如：
library(plyr)
library(ggplot2)
library(gtable)
library(grid)
library(gridExtra)
library(ClustOfVar)
library(ggdendro)
fit = hclustvar(X.quanti = mtcars)
labels = cutree(fit,k = 5)
labelx = data.frame(Names=names(labels),group = paste("Group",as.vector(labels)),num=as.vector(labels))
p1 = ggdendrogram(as.dendrogram(fit), rotate=TRUE)
df2<-data.frame(cluster=cutree(fit, k =5), states=factor(fit$labels,levels=fit$labels[fit$order]))
df3<-ddply(df2,.(cluster),summarise,pos=mean(as.numeric(states)))
p2 = ggplot(df2,aes(states,y=1,fill=factor(cluster)))+geom_tile()+
scale_y_continuous(expand=c(0,0))+
theme(axis.title=element_blank(),
axis.ticks=element_blank(),
axis.text=element_blank(),
legend.position="none")+coord_flip()+
geom_text(data=df3,aes(x=pos,label=cluster))
gp1<-ggplotGrob(p1)
gp2<-ggplotGrob(p2)
maxHeight = grid::unit.pmax(gp1$heights[2:5], gp2$heights[2:5])
gp1$heights[2:5] <- as.list(maxHeight)
gp2$heights[2:5] <- as.list(maxHeight)
grid.arrange(gp2, gp1, ncol=2,widths=c(1/6,5/6))
当列数较多时，会出现另一个问题。即色块部分的高度与树状图的高度不匹配。
library(ClustOfVar)
library(ggdendro)
X = data.frame(mtcars,mtcars,mtcars,mtcars,mtcars,mtcars)
fit = hclustvar(X.quanti = X)
labels = cutree(fit,k = 5)
labelx = data.frame(Names=names(labels),group = paste("Group",as.vector(labels)),num=as.vector(labels))
p1 = ggdendrogram(as.dendrogram(fit), rotate=TRUE)
df2<-data.frame(cluster=cutree(fit, k =5), states=factor(fit$labels,levels=fit$labels[fit$order]))
df3<-ddply(df2,.(cluster),summarise,pos=mean(as.numeric(states)))
p2 = ggplot(df2,aes(states,y=1,fill=factor(cluster)))+geom_tile()+
scale_y_continuous(expand=c(0,0))+
theme(axis.title=element_blank(),
axis.ticks=element_blank(),
axis.text=element_blank(),
legend.position="none")+coord_flip()+
geom_text(data=df3,aes(x=pos,label=cluster))
gp1<-ggplotGrob(p1)
gp2<-ggplotGrob(p2)
maxHeight = grid::unit.pmax(gp1$heights[2:5], gp2$heights[2:5])
gp1$heights[2:5] <- as.list(maxHeight)
gp2$heights[2:5] <- as.list(maxHeight)
grid.arrange(gp2, gp1, ncol=2,widths=c(1/6,5/6))
@Sandy Muspratt 实际上为此提供了一个很好的解决方案如果我们已经将 R 升级到版本 3.3.1。
但是由于我无法更改部署在公司服务器中的 R 版本，我想知道是否有任何其他解决方法可以使这两个部分保持一致。
据我所知，您的代码并没有大错。问题是当您合并这两个图时，您正试图将连续比例与离散比例相匹配。此外，似乎 ggdendrogram() 向 y-axis 添加了额外的 space。
library(plyr)
library(ggplot2)
library(gtable)
library(grid)
library(gridExtra)
library(ClustOfVar)
library(ggdendro)
# Data
X = data.frame(mtcars,mtcars,mtcars,mtcars,mtcars,mtcars)
# Cluster analysis
fit = hclustvar(X.quanti = X)
# Labels data frames
df2 <- data.frame(cluster = cutree(fit, k =5),
states = factor(fit$labels, levels = fit$labels[fit$order]))
df3 <- ddply(df2, .(cluster), summarise, pos = mean(as.numeric(states)))
# Dendrogram
# scale_x_continuous() for p1 should match scale_x_discrete() from p2
# scale_x_continuous strips off the labels. I grab them from df2
# scale _y_continuous() puts a little space between the labels and the dendrogram
p1 <- ggdendrogram(as.dendrogram(fit), rotate = TRUE) +
scale_x_continuous(expand = c(0, 0.5), labels = levels(df2$states), breaks = 1:length(df2$states)) +
scale_y_continuous(expand = c(0.02, 0))
# Tiles and labels
p2 <- ggplot(df2,aes(states, y = 1, fill = factor(cluster))) +
geom_tile() +
scale_y_continuous(expand = c(0, 0)) +
scale_x_discrete(expand = c(0, 0)) +
geom_text(data = df3, aes(x = pos, label = cluster)) +
coord_flip() +
theme(axis.title = element_blank(),
axis.ticks = element_blank(),
axis.text = element_blank(),
legend.position = "none")
# Get the ggplot grobs
gp1 <- ggplotGrob(p1)
gp2 <- ggplotGrob(p2)
# Make sure the heights match
maxHeight <- unit.pmax(gp1$heights, gp2$heights)
gp1$heights <- as.list(maxHeight)
gp2$heights <- as.list(maxHeight)
# Combine the two plots
grid.arrange(gp2, gp1, ncol = 2,widths = c(1/6, 5/6))
想法是结合 R 包 ClustOfVar 和 ggdendro 给出变量聚类的可视化摘要。
当数据列数较少时，除了有未覆盖的区域（如下图圈出）外，效果还不错。使用 mtcars 例如：
library(plyr)
library(ggplot2)
library(gtable)
library(grid)
library(gridExtra)
library(ClustOfVar)
library(ggdendro)
fit = hclustvar(X.quanti = mtcars)
labels = cutree(fit,k = 5)
labelx = data.frame(Names=names(labels),group = paste("Group",as.vector(labels)),num=as.vector(labels))
p1 = ggdendrogram(as.dendrogram(fit), rotate=TRUE)
df2<-data.frame(cluster=cutree(fit, k =5), states=factor(fit$labels,levels=fit$labels[fit$order]))
df3<-ddply(df2,.(cluster),summarise,pos=mean(as.numeric(states)))
p2 = ggplot(df2,aes(states,y=1,fill=factor(cluster)))+geom_tile()+
scale_y_continuous(expand=c(0,0))+
theme(axis.title=element_blank(),
axis.ticks=element_blank(),
axis.text=element_blank(),
legend.position="none")+coord_flip()+
geom_text(data=df3,aes(x=pos,label=cluster))
gp1<-ggplotGrob(p1)
gp2<-ggplotGrob(p2)
maxHeight = grid::unit.pmax(gp1$heights[2:5], gp2$heights[2:5])
gp1$heights[2:5] <- as.list(maxHeight)
gp2$heights[2:5] <- as.list(maxHeight)
grid.arrange(gp2, gp1, ncol=2,widths=c(1/6,5/6))
当列数较多时，会出现另一个问题。即色块部分的高度与树状图的高度不匹配。
library(ClustOfVar)
library(ggdendro)
X = data.frame(mtcars,mtcars,mtcars,mtcars,mtcars,mtcars)
fit = hclustvar(X.quanti = X)
labels = cutree(fit,k = 5)
labelx = data.frame(Names=names(labels),group = paste("Group",as.vector(labels)),num=as.vector(labels))
p1 = ggdendrogram(as.dendrogram(fit), rotate=TRUE)
df2<-data.frame(cluster=cutree(fit, k =5), states=factor(fit$labels,levels=fit$labels[fit$order]))
df3<-ddply(df2,.(cluster),summarise,pos=mean(as.numeric(states)))
p2 = ggplot(df2,aes(states,y=1,fill=factor(cluster)))+geom_tile()+
scale_y_continuous(expand=c(0,0))+
theme(axis.title=element_blank(),
axis.ticks=element_blank(),
axis.text=element_blank(),
legend.position="none")+coord_flip()+
geom_text(data=df3,aes(x=pos,label=cluster))
gp1<-ggplotGrob(p1)
gp2<-ggplotGrob(p2)
maxHeight = grid::unit.pmax(gp1$heights[2:5], gp2$heights[2:5])
gp1$heights[2:5] <- as.list(maxHeight)
gp2$heights[2:5] <- as.list(maxHeight)
grid.arrange(gp2, gp1, ncol=2,widths=c(1/6,5/6))
@Sandy Muspratt 实际上为此提供了一个很好的解决方案如果我们已经将 R 升级到版本 3.3.1。
但是由于我无法更改部署在公司服务器中的 R 版本，我想知道是否有任何其他解决方法可以使这两个部分保持一致。
据我所知，您的代码并没有大错。问题是当您合并这两个图时，您正试图将连续比例与离散比例相匹配。此外，似乎 ggdendrogram() 向 y-axis 添加了额外的 space。
library(plyr)
library(ggplot2)
library(gtable)
library(grid)
library(gridExtra)
library(ClustOfVar)
library(ggdendro)
# Data
X = data.frame(mtcars,mtcars,mtcars,mtcars,mtcars,mtcars)
# Cluster analysis
fit = hclustvar(X.quanti = X)
# Labels data frames
df2 <- data.frame(cluster = cutree(fit, k =5),
states = factor(fit$labels, levels = fit$labels[fit$order]))
df3 <- ddply(df2, .(cluster), summarise, pos = mean(as.numeric(states)))
# Dendrogram
# scale_x_continuous() for p1 should match scale_x_discrete() from p2
# scale_x_continuous strips off the labels. I grab them from df2
# scale _y_continuous() puts a little space between the labels and the dendrogram
p1 <- ggdendrogram(as.dendrogram(fit), rotate = TRUE) +
scale_x_continuous(expand = c(0, 0.5), labels = levels(df2$states), breaks = 1:length(df2$states)) +
scale_y_continuous(expand = c(0.02, 0))
# Tiles and labels
p2 <- ggplot(df2,aes(states, y = 1, fill = factor(cluster))) +
geom_tile() +
scale_y_continuous(expand = c(0, 0)) +
scale_x_discrete(expand = c(0, 0)) +
geom_text(data = df3, aes(x = pos, label = cluster)) +
coord_flip() +
theme(axis.title = element_blank(),
axis.ticks = element_blank(),
axis.text = element_blank(),
legend.position = "none")
# Get the ggplot grobs
gp1 <- ggplotGrob(p1)
gp2 <- ggplotGrob(p2)
# Make sure the heights match
maxHeight <- unit.pmax(gp1$heights, gp2$heights)
gp1$heights <- as.list(maxHeight)
gp2$heights <- as.list(maxHeight)
# Combine the two plots
grid.arrange(gp2, gp1, ncol = 2,widths = c(1/6, 5/6))