[Feature Request]: `values(keep_na = logical(1))` to handle `NA`

[gogonzo]

Feature description

Related to very old issue present in des also: insightsengineering/teal.transform#30

We already found a quite nice solution in teal_slice(keep_na). In teal.slice when column contains any NA, checkbox input is automatically added to the filter-state-card. We expect similar in pick_ui

Consider following example

data <- within(teal.data::teal_data(), {
  iris <- iris
  mtcars <- mtcars
  iris$Species[1:5] <- NA 
})

app <- init(
  data = data,
  modules = modules(
    modules(
      label = "Testing modules",
      tm_merge(
        label = "non adam",
        picks = list(
          a = picks(
            datasets("iris", "iris"),
            variables(
              choices = c("Sepal.Length", "Species"),
              selected = "Species"
            ),
            values()
          )
        )
      )
    )
  )
)

shinyApp(app$ui, app$server, enableBookmarking = "server")

1. how NA should be included in choices/selected?

values(choices, selected, keep_na = TRUE/FALSE)
# adds keep_na to the attributes of the `pick` object

2. how NA should be displayed handled in the pick_ui/srv?

• checkbox-input should be added below pickerInput (or rangeSlider or other input related with class of the column)
• changing input should update attribute in the object

3. how NA should be handled in merge_srv

Reactive keep_na should be delivered somehow to merge_srv > .qenv_merge > .merge_expr > .call_dplyr_filter and handled respectively.

[llrs-roche]

Perhaps this is a separate issue but it occurred to me how would we handle cases where there are no column names? They aren't NA: colnames(matrix(ncol = 5)), but on a data.frame they get name V1, so it might not be a big issue.

This would affect variables() if we want to address it.

[osenan]

So the scenario that you are describing @llrs-roche is this example app:

library(shiny)
library(teal.picks)

mydata <- matrix(1:12, nrow = 3) |>
  as.data.frame()

my_picks <- teal.picks::picks(
  teal.picks::datasets("mydata"),
  teal.picks::variables()
)

ui <- fluidPage(
  teal.picks::picks_ui("my_picks", picks = my_picks)
)

server <- function(input, output, session) {
  teal.picks::picks_srv(
    "my_picks",
    picks = my_picks,
    data = reactive(list(mydata = mydata))
  )
}

shinyApp(ui, server)

I think that the package is already responding to what we should expect. We cannot create a picks object the dataset is not a dataframe. When we convert a matrix with unnamed columns it will create the generic column names "V1" ... and this will appear as variables. Developers still can use filter is_categorical or others to select those variables of interest based on class, not on column name which is logical as we do not have column names to filter variables.