ggplot 在 shiny 的 eventReactive() 内无法正常工作

Question

I wasted hours to find out why my plot is automatically updating itself when I change inputs while it was supposed to wait for the Run button but it simply ignored that step and I ended up finally finding ggplot as the trouble maker:!! This is my minimal code:

library(ggplot2)
library(tidyverse)

varnames <- names(cars)

ui <- fluidPage(
  sidebarLayout(
    sidebarPanel(
      fluidRow(
        column(
          width = 12,

          # Variables Inputs:
          varSelectInput("variables", "Select Input Variables", cars, multiple = TRUE),
          selectizeInput("outvar", "Select Output Variable", choices = varnames, "speed", multiple = F),

          # Run Button
          actionButton(inputId = "run", label = "Run")
        )
      )
    ),

    # Main panel for displaying outputs ----
    mainPanel(
      plotOutput("plot")
    )
  )
)

server <- function(input, output, session) {
  
  df <- reactive({
    cars %>% dplyr::select(!!!input$variables, input$outvar)
  })


  plt <- eventReactive(input$run, {
    
    #Just creating lm formula
    current_formula <- paste0(input$outvar, " ~ ", paste0(input$variables, collapse = " + "))
    current_formula <- as.formula(current_formula)
    #Fitting lm
    fit <- lm(current_formula, data = df())
    pred <- predict(fit, newdata = df())

    #Plotting
    ggplot(df(), aes(df()[, input$outvar], pred)) +
      labs(x = "Observed", y = "Predicted") +
      geom_point() +
      theme_bw()

     #plot(df()[, input$outvar], pred)       #This one works fine!!!!
  })


  output$plot <- renderPlot({
     plt()
  })
}

# Run the application
shinyApp(ui = ui, server = server)

If you run this, you'll notice that ggplot doesn't care anymore about the Run button after the 1st run and it keeps updating as you change the inputs,! However, if you use the simple base plot function (which I put in a comment in the code) there wouldn't be any problems and that works just fine. Sadly I need ggplot in my app because base plot is ugly. I am seeing suggestion for using isolate() to solve this issue but I have no clue where isolate() should be put to fix my problem also it doesn't make sense to use isolate() when base plot function works fine without it and it's the ggplot that makes the problem. Any explanation would be appreciated.

Answer 1

The issue is that ggplot aesthetics are lazy evaluated. You really want to put symbols into the aes() rather that reactive data values. Change your plotting code to

ggplot(df(), aes(.data[[input$outvar]], pred)) +
  labs(x = "Observed", y = "Predicted") +
  geom_point() +
  theme_bw()

With ggplot you use the .data pronoun to access the current data source rather than trigger the reactive df() object again.