了解 3 维 kmeans 图 [关闭]

Question

以下代码生成此图：

当对二维项目进行聚类时，每个聚类都有一个质心，但为什么这些图没有生成质心？

每组图是否生成其他两项的 kmeans 集群？因此，例如在从左到右的第一行中，“google”是标签，正在为“so”和“test”生成kmeans，这是正确的吗？

cells = c(1,1,1,
          1,0,1,
          1,0,1,
          1,0,0,
          1,1,1,
          0,1,0,
          0,1,1,
          1,1,0,
          0,0,1,
          0,0,0,
          1,1,1,
          1,1,0,
          1,0,1,
          1,1,0,
          1,0,1,
          1,1,0,
          1,0,1,
          1,1,0,
          1,0,1,
          1,1,0,
          1,0,1,
          1,1,0,
          1,0,1,
          1,1,0)
rnames = c("a1","a2","a3","a4","a5","a6","a7","a8","a9","10","11","12","13","14","15","16","17","18","19","20","21","22","23","24")
cnames = c("google","so","test")
x <- matrix(cells, nrow=24, ncol=3, byrow=TRUE, dimnames=list(rnames, cnames))
# run K-Means
km <- kmeans(x, 8, 5)
# print components of km
print(km)
# plot clusters
plot(x, col = km$cluster)
# plot centers
pairs(jitter(x), col = cl$cluster)

Answer 1

因为您没有绘制质心。在您的earlier question 中，质心是通过以下命令绘制的：

points(cl$centers, col = 1:5, pch = 8, cex = 2)

这会将每个质心的点添加到plot 函数生成的图中。如果您尝试使用pairs() 执行此操作，它将无法正常工作。但是你甚至没有在你发布的代码中尝试这个，所以我不确定你为什么希望看到绘制的质心。

不幸的是，将点添加到pairs() 图中是一个手动过程。您可以使用pairs() 函数的panel、lower.panel 和upper.panel 参数来准确指定要为每对向量绘制的内容。在这里，我指定子方法在顶部面板中正常绘制点，并在下部面板中绘制具有质心的点。

# I use the variable name "x" elsewhere, 
# renaming it here explicitly for clarity  
x.mat=x

# I moved the "jitter" into this submethod, so you won't see it
# in the main 'pairs()' call. I needed to do this to identify the source
# column the data came from in low.panelfun.
up.panelfun <- function(x,y,clust=cl$cluster,...){
  # this plots the main pairs plot
  sapply(unique(clust), function(c){ points(jitter(x[clust==c]),jitter(y[clust==c]), col=c)}) 
}

low.panelfun <- function(x,y,clust=cl$cluster,...){
  # this plots the main pairs plot
  up.panelfun(x,y,clust)

  # this finds the appropriate column the panel is related
  # to and plots the centroids.
  xi=which(length(x)==apply(x.mat, 2, function(v){sum(v==x)}))
  yi=which(length(y)==apply(x.mat, 2, function(v){sum(v==y)}))
  points(cl$centers[xi,],cl$centers[yi,], col = 1:5, pch = 8, cex = 2)
}

pairs(x.mat, col = cl$cluster
      ,lower.panel=low.panelfun
      ,upper.panel=up.panelfun
)

由于您的数据集非常小，我发现通过将结果复制几次以使集群更加明显来放大数据很有用：

# amplify clusters by replicating data a few times
pairs(rbind(x.mat, x.mat, x.mat, x.mat), col = cl$cluster
      ,lower.panel=low.panelfun
      ,upper.panel=up.panelfun
)

考虑到这需要做的所有额外工作，并且您实际上只需要三个图，为每对变量构建单独的 plot();points() 调用可能会更容易。