Querychat drops factor levels? - affects charts

[garrettgman]

When I use querychat, charts that display factors begin to order levels alphabetically. This makes me think that querychat drops factor levels when processing the data. Here is a toy example.

A shiny app

library(shiny)
library(bslib)
library(forcats)
library(ggplot2)
library(dplyr)

relig_choices <- c("All", levels(gss_cat$relig))

ui <- page_sidebar(
  title = "GSS Marital Status",
  sidebar = sidebar(
    selectInput("relig", "Religion", choices = relig_choices, selected = "All")
  ),
  plotOutput("marital_plot")
)

server <- function(input, output, session) {
  filtered_data <- reactive({
    if (input$relig == "All") {
      gss_cat
    } else {
      filter(gss_cat, relig == input$relig)
    }
  })

  output$marital_plot <- renderPlot({
    ggplot(filtered_data(), aes(x = marital, fill = marital)) +
      geom_bar() +
      theme(
        axis.text  = element_text(size = 16),
        axis.title = element_text(size = 18)
      )
  })
}

shinyApp(ui, server)

The querychat version

library(shiny)
library(bslib)
library(forcats)
library(ggplot2)
library(dplyr)
library(querychat)

qc <- QueryChat$new(gss_cat, greeting = "Hello")


ui <- page_sidebar(
  title = "GSS Marital Status",
  sidebar = qc$sidebar(),
  plotOutput("marital_plot")
)

server <- function(input, output, session) {

  qc_vals <- qc$server()

  filtered_data <- reactive({
    qc_vals$df()
  })

  output$marital_plot <- renderPlot({
    ggplot(filtered_data(), aes(x = marital)) +
      geom_bar()
  })
}

shinyApp(ui, server)

[gadenbuie]

Thanks for the clear reprex, Garrett!

You're right, and it's expected (if frustrating). querychat runs queries against an in-memory DuckDB or SQLite table, and neither has a concept of R factors. Results come back as plain character columns with the level ordering gone.

As a workaround, you can re-apply the factor levels from the original data after calling $df():

filtered_data <- reactive({
  df <- qc_vals$df()
  df$marital <- factor(df$marital, levels = levels(gss_cat$marital))
  df
})

querychat could potentially detect factor columns and restore their levels automatically, but there's some nuance to sort out first. We'd want to think it through before adding it.