Cluster Analysis"
In CLUSTShiny: Interactive Document for Working with Cluster Analysis

knitr::opts_chunk$set(echo = TRUE)
library(dplyr)
library(datasets)
library(psycho)

Hierarchical and K-means Clustering

The distance measure used is Euclidean in case of both Hierarchical and K-means(Non-hierarchical) Clustering. The method(algorithm) used is Ward's method for hierarchical clustering and k-means algorithm for non-hierarchical clustering. Hierarchical Clustering is for exploration .So is Silhouette plot for detecting outliers. K-means Clustering is meant for confirmation. K-means is done usually after hierarchical clustering. There would be a difference between the cluster sizes from the two algorithms. Final Clusters and membership is usually taken from k-means clustering.

sidebarPanel(
  checkboxInput("ex","Uncheck for using your own file",value = TRUE),
  fileInput("file", "Upload the *.csv file with headers"),
  numericInput("ncluster","Enter the number of clusters",value = 2),

  checkboxInput("std","STANDARDIZE",value = FALSE),
  downloadButton("downloaddata", "Download Dataset"),
  downloadButton("downloadPlot", "Download Plot"),
  downloadButton("downloadPlot2", "Download Dendrogram"),
  downloadButton("downloadPlot3", "Download Silhouette"),
  uiOutput("vx")


)
mainPanel(
   tabsetPanel(type = "tab",
                tabPanel("Hierarchical Clustering Visualization",plotOutput("HV")),
                tabPanel("Detecting outliers : Silhouette Plot(hierarchical)",plotOutput("SP")),
                tabPanel("K- means Clustering", verbatimTextOutput("AD") ),
                tabPanel("Cluster Visualization(k-means)", plotOutput("MV") )

                ),

   h6("", tags$img(src ="K.JPG", height= 400, width=400))
)
output$AD<-renderPrint({
   if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else{
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }
     ds = select(data,input$variablex)
     if(input$std == TRUE)
     {
       ds = standardize(ds)
     }
     set.seed(1)
     fitkmeans =  kmeans(ds,centers = input$ncluster )

     cat(sprintf("\nThe Cluster centres are as follows:\n"))
     print(fitkmeans$centers)
     cat(sprintf("\nThe Cluster sizes are as follows:\n"))
     cat(sprintf("%d",fitkmeans$size))
})

output$MV<-renderPlot({ 
  if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else{
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }
     ds = data

     ds = select(ds,input$variablex)

     if(input$std == TRUE)
     {
       ds = standardize(ds)
     }
     set.seed(1)
     fitkmeans =  kmeans(ds,centers = input$ncluster )
     ds$cluster = as.factor(fitkmeans$cluster)




    if(ncol(ds)>2 && input$ncluster >1)
    {  klaR:: partimat(formula = cluster~.,data = ds,method="lda")}
   else
   { 
     if(input$ncluster>=1 && ncol(ds)==2)
   {  attach(ds)
     boxplot(get(input$variablex)~cluster,col ="red",ylab = input$variablex,xlab = "cluster")
     }
   } 

#     } 



})


output$vx <- renderUI({


    if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else
     {
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }

     ds = select_if(data,is.numeric)

checkboxGroupInput("variablex","Select the set of quantitative variables",choices = colnames(ds),selected = colnames(ds) )



  })

datasetInput1 <- reactive({

    if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else
     {
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }

     ds = data
     ds = select(ds,input$variablex)

     if(input$std == TRUE)
     {
       ds = standardize(ds)
     }
     set.seed(1)
     fitkmeans =  kmeans(ds,centers = input$ncluster )
     data$cluster = as.factor(fitkmeans$cluster)

     data = data

})
output$downloaddata <- downloadHandler(
    filename = function() {
      filetitle = paste("dataset")
      paste(filetitle, ".csv", sep = "")
    },
    content = function(file) {

      write.csv(datasetInput1(), file, row.names = FALSE)
    }
  )
output$downloadPlot<- downloadHandler(
    filename = function() {
      paste("Discriminantplot", ".png", sep = "")
    },
    content = function(file) {
      png(file)
      if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else{
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }
     ds = data

     ds = select(ds,input$variablex)

     if(input$std == TRUE)
     {
       ds = standardize(ds)
     }
     set.seed(1)
     fitkmeans =  kmeans(ds,centers = input$ncluster )
     ds$cluster = as.factor(fitkmeans$cluster)




    if(ncol(ds)>2 && input$ncluster >1)
    {  klaR:: partimat(formula = cluster~.,data = ds,method="lda")}
   else
   { 
     if(input$ncluster>=1 && ncol(ds)==2)
   {  attach(ds)
     boxplot(get(input$variablex)~cluster,col ="red",ylab = input$variablex,xlab = "cluster")
     }
   } 

#     } 




      dev.off()
    })

output$HV<-renderPlot({
   if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else{
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }
     ds = select(data,input$variablex)
     if(input$std == TRUE)
     {
       ds = standardize(ds)
     }

d = dist(ds,method = "euclidean") # Measure of similarity
fit = hclust(d,method = "ward.D") # Method of using similarity measure
plot(fit)
rect.hclust(fit,k=input$ncluster,border = "red")
#groups = cutree(fit,k=input$ncluster)
#plot(cluster::silhouette(groups,d))

})

output$downloadPlot2<- downloadHandler(
    filename = function() {
      paste("Dendogram", ".png", sep = "")
    },
    content = function(file) {
      png(file)

if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else{
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }
     ds = select(data,input$variablex)
     if(input$std == TRUE)
     {
       ds = standardize(ds)
     }

d = dist(ds,method = "euclidean") # Measure of similarity
fit = hclust(d,method = "ward.D") # Method of using similarity measure
plot(fit)
rect.hclust(fit,k=input$ncluster,border = "red")



      dev.off()
    })

output$SP<-renderPlot({
   if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else{
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }
     ds = select(data,input$variablex)
     if(input$std == TRUE)
     {
       ds = standardize(ds)
     }

d = dist(ds,method = "euclidean") # Measure of similarity
fit = hclust(d,method = "ward.D") # Method of using similarity measure
#plot(fit)
#rect.hclust(fit,k=input$ncluster,border = "red")
groups = cutree(fit,k=input$ncluster)
plot(cluster::silhouette(groups,d))

})

output$downloadPlot3<- downloadHandler(
    filename = function() {
      paste("Silhouette", ".png", sep = "")
    },
    content = function(file) {
      png(file)

if(input$ex == TRUE)
     {data("iris")
      data = iris}
     else{
  file1 = input$file
     if(is.null(file1)){return()}

     data =  read.table(file =  file1$datapath,sep =",",header = TRUE)
     if(is.null(data())){return()}
     }
     ds = select(data,input$variablex)
     if(input$std == TRUE)
     {
       ds = standardize(ds)
     }

d = dist(ds,method = "euclidean") # Measure of similarity
fit = hclust(d,method = "ward.D") # Method of using similarity measure
#plot(fit)
#rect.hclust(fit,k=input$ncluster,border = "red")
 groups = cutree(fit,k=input$ncluster)
plot(cluster::silhouette(groups,d))    


      dev.off()
    })