提取集群信息并合并结果

问题描述

我正在尝试针对不同数量的群集k在一系列不同矩阵上运行聚类算法，并为每次运行提取一些信息。

第一段代码生成了相异矩阵列表

library(tidyverse)
library(cluster)
library(rje)   

dat=mtcars[,1:3]
v_names=names(dat)

combos=rje::powerSet(v_names)

combos=combos[lengths(combos)>1]


df_list=list()

for (i in seq_along(combos)){

  df_list[[i]]=dat[combos[[i]]]
  
  }
gower_ls=lapply(df_list,daisy,metric="gower")

这是我遇到问题的代码部分

set.seed(4)
model_num <-c(NA)    
sil_width <-c(NA)    
min_sil<-c(NA)    
mincluster<-c(NA)    
k_clusters <-c(NA)    
lowest_sil <-c(NA)    
maxcluster <-c(NA)    
model_vars <- c(NA)

clust_4=lapply(gower_ls,pam,diss=TRUE,k=4)    

for(m in 1:length(clust_4)){
  
  sil_width[m] <-clust_4[[m]][7]$silinfo$avg.width
  
  min_sil[m] <- min(clust_4[[m]][7]$silinfo$clus.avg.widths)
  
  mincluster[m] <-min(clust_4[[m]][6]$clusinfo[,1])
  
  maxcluster[m] <-max(clust_4[[m]][6]$clusinfo[,1])
  
  k_clusters[m]<- nrow(clust_4[[m]][6]$clusinfo)
  
  lowest_sil[m]<-min(clust_4[[m]][7]$silinfo$widths)
  
  model_num[m] <-m
 }    

colresults_4=as.data.frame(cbind( sil_width,min_sil,mincluster,maxcluster,k_clusters,model_num,lowest_sil))

如何将这段代码转换为在给定的k范围内运行？我尝试了嵌套循环，但是无法正确编码。这是k= 4:6的理想结果，谢谢。

structure(list(sil_width = c(0.766467312788453,0.543226669407726,0.765018469447229,0.705326458357873,0.698351173575526,0.480565022092276,0.753366365875066,0.644345251543097,0.699437672202048,0.430310752506775,0.678224885117295,0.576411380463116),min_sil = c(0.539324315243191,0.508330909368204,0.637090842537915,0.622120627356455,0.539324315243191,0.334047777245833,0.430814518122641,0.568591550281139,0.295113900268025,0.19040716086259),mincluster = c(5,3,4,5,2,3),maxcluster = c(14,12,11,14,10,9,6,7,7),k_clusters = c(4,6),model_num = c(1,1,4),lowest_sil = c(-0.0726256983240229,0.0367238314801671,0.308069836672298,0.294247157041013,-0.0726256983240229,-0.122804288130541,-0.317748917748917,0.218164082936686,-0.224849074123824,-0.459909237820881)),row.names = c(NA,-12L
),class = "data.frame")

解决方法

我编写了一个函数clus_func来提取集群信息，然后使用cross2包中的map2和purrr提出了一个解决方案：

    library(tidyverse)
    library(cluster)
    library(rje)   
    
    dat=mtcars[,1:3]
    v_names=names(dat)
    
    combos=rje::powerSet(v_names)
    
    combos=combos[lengths(combos)>1]
    
    clus_func=function(x,k){
  
    clust=pam(x,k,diss=TRUE)
    clust_stats=as.data.frame(cbind(
    avg_sil_width=clust$silinfo$avg.width,min_clus_width=min(clust$silinfo$clus.avg.widths),min_individual_sil=min(clust$silinfo$widths[,3]),max_individual_sil=max(clust$silinfo$widths[,mincluster= min(clust$clusinfo[,1]),maxcluster= max(clust$clusinfo[,num_k=max(clust$clustering)  ))       
  
     } 

    
    df_list=list()
    
    for (i in seq_along(combos)){
    
      df_list[[i]]=dat[combos[[i]]]
      
      }
    gower_ls=lapply(df_list,daisy,metric="gower")
        begin_k=4
        end_k=6
        cross_list=cross2(gower_ls,begin_k:end_k)
        
        k=c(NA)
        for(i in 1:length(cross_list)){ k[i]=cross_list[[i]][2]}
        diss=c(NA)
        for(i in 1:length(cross_list)){ diss[i]=cross_list[[i]][1]}
        
        model_stats=map2(diss,clus_func)
        model_stats=rbindlist(model_stats)