how do I plot my scatterplot on a map? I managed to plot my scatterplot, however I wanted it to be plotted on a map. I believe that an option is to use the leaflet package, since I have the Latitude and Longitude coordinates, but I don't know how to use it. Please, if you have other options feel free.
Could you help me with this problem ?? The executable code is below.
Thank you very much!
library(shiny)
library(ggplot2)
library(rdist)
library(geosphere)
library(kableExtra)
library(readxl)
library(tidyverse)
library(DT)
#database
df<-structure(list(Properties = c(1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35), Latitude = c(-23.8, -23.8, -23.9, -23.9, -23.9, -23.9, -23.9, -23.9, -23.9, -23.9, -23.9, -23.9, -23.9, -23.9,
+ -23.9, -23.9, -23.9, -23.9, -23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9,-23.9), Longitude = c(-49.6, -49.6, -49.6, -49.6, -49.6, -49.6, -49.6, -49.6, -49.6, -49.6, -49.7,
+ -49.7, -49.7, -49.7, -49.7, -49.6, -49.6, -49.6, -49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6,-49.6), Waste = c(526, 350, 526, 469, 285, 175, 175, 350, 350, 175, 350, 175, 175, 364,
+ 175, 175, 350, 45.5, 54.6,350,350,350,350,350,350,350,350,350,350,350,350,350,350,350,350)), class = "data.frame", row.names = c(NA, -35L))
function.clustering<-function(df,k,Filter1,Filter2){
if (Filter1==2){
Q1<-matrix(quantile(df$Waste, probs = 0.25))
Q3<-matrix(quantile(df$Waste, probs = 0.75))
L<-Q1-1.5*(Q3-Q1)
S<-Q3+1.5*(Q3-Q1)
df_1<-subset(df,Waste>L[1])
df<-subset(df_1,Waste<S[1])
}
#cluster
coordinates<-df[c("Latitude","Longitude")]
d<-as.dist(distm(coordinates[,2:1]))
fit.average<-hclust(d,method="average")
#Number of clusters
clusters<-cutree(fit.average, k)
nclusters<-matrix(table(clusters))
df$cluster <- clusters
#Localization
center_mass<-matrix(nrow=k,ncol=2)
for(i in 1:k){
center_mass[i,]<-c(weighted.mean(subset(df,cluster==i)$Latitude,subset(df,cluster==i)$Waste),
weighted.mean(subset(df,cluster==i)$Longitude,subset(df,cluster==i)$Waste))}
coordinates$cluster<-clusters
center_mass<-cbind(center_mass,matrix(c(1:k),ncol=1))
#Coverage
coverage<-matrix(nrow=k,ncol=1)
for(i in 1:k){
aux_dist<-distm(rbind(subset(coordinates,cluster==i),center_mass[i,])[,2:1])
coverage[i,]<-max(aux_dist[nclusters[i,1]+1,])}
coverage<-cbind(coverage,matrix(c(1:k),ncol=1))
colnames(coverage)<-c("Coverage_meters","cluster")
#Sum of Waste from clusters
sum_waste<-matrix(nrow=k,ncol=1)
for(i in 1:k){
sum_waste[i,]<-sum(subset(df,cluster==i)["Waste"])
}
sum_waste<-cbind(sum_waste,matrix(c(1:k),ncol=1))
colnames(sum_waste)<-c("Potential_Waste_m3","cluster")
#Output table
data_table <- Reduce(merge, list(df, coverage, sum_waste))
data_table <- data_table[order(data_table$cluster, as.numeric(data_table$Properties)),]
data_table_1 <- aggregate(. ~ cluster + Coverage_meters + Potential_Waste_m3, data_table[,c(1,7,6,2)], toString)
#Scatter Plot
suppressPackageStartupMessages(library(ggplot2))
df1<-as.data.frame(center_mass)
colnames(df1) <-c("Latitude", "Longitude", "cluster")
g<-ggplot(data=df, aes(x=Longitude, y=Latitude, color=factor(clusters))) + geom_point(aes(x=Longitude, y=Latitude), size = 4)
Centro_View<- g + geom_text(data=df, mapping=aes(x=eval(Longitude), y=eval(Latitude), label=Waste), size=3, hjust=-0.1)+ geom_point(data=df1, mapping=aes(Longitude, Latitude), color= "green", size=4) + geom_text(data=df1, mapping = aes(x=Longitude, y=Latitude, label = 1:k), color = "black", size = 4)
plotGD<-print(Centro_View + ggtitle("Scatter Plot") + theme(plot.title = element_text(hjust = 0.5)))
return(list(
"Data" = data_table_1,
"Plot" = plotGD
))
}
ui <- bootstrapPage(
navbarPage(theme = shinytheme("flatly"), collapsible = TRUE,
"Clustering",
tabPanel("General Solution",
sidebarLayout(
sidebarPanel(
radioButtons("filtro1", h3("Select properties"),
choices = list("All properties" = 1,
"Exclude properties" = 2),
selected = 1),
tags$b(h5("(a) Choose other filters")),
tags$b(h5("(b) Choose clusters")),
sliderInput("Slider", h5(""),
min = 2, max = 8, value = 5)
),
mainPanel(
tabsetPanel(
tabPanel("Solution", plotOutput("ScatterPlot"))))
))))
server <- function(input, output, session) {
Modelclustering<-reactive(function.clustering(df,input$Slider,1,1))
output$ScatterPlot <- renderPlot({
Modelclustering()[[2]]
})
observeEvent(input$Slider,{
updateSelectInput(session,'select',
choices=unique(df[df==input$Slider]))
})
}
shinyApp(ui = ui, server = server)
Thank you very much!
See Question&Answers more detail:
os