Transform Input Data

Introduction

teal version 0.16 introduced a new, optional argument in teal::module, transformators. This argument accepts a list of teal_transform_module objects, which are created using the teal_transform_module() function. teal_transform_module() takes ui and server arguments to create a shiny module that encodes data transformations.

When transformators are passed to a module, teal will execute data transformations when that module is loaded as well as whenever the original data changes. The transformations are applied to the data before it reaches the module.

The ui elements of the transform module will be added to the filter panel, while the server function provides the data manipulation logic.

This vignette describes how to manage custom data transformations in teal apps.

Transforming teal_data

In this vignette we will focus on using the teal_transform_module for transforming the input data using the transformators argument in teal::module function.

Creating a data transformation module

Let us initialize a simple teal app by providing iris and mtcars as input datasets.

library(teal)

data <- within(teal_data(), {
  iris <- iris
  mtcars <- mtcars
})

app <- init(
  data = data,
  modules = example_module()
)

if (interactive()) {
  shinyApp(app$ui, app$server)
}

Single transformator

Now let us create a simple teal_transform_module that returns the first n number of rows of iris based on user input.

We will achieve this by creating a UI function with a numericInput for the user to specify the number of rows to be displayed. The server function will take a reactive expression holding data as argument and return a reactive expression holding transformed data.

Note: It is recommended to return reactive() with teal_data() in server code of a teal_transform_module as this is more robust for maintaining the reactivity of Shiny. If you are planning on using eventReactive() in the server, the event should include data() (example eventReactive(list(input$a, data()), {...})). More in this discussion.

data <- within(teal_data(), {
  iris <- iris
  mtcars <- mtcars
})

transformator_iris <- teal_transform_module(
  label = "Custom transformator for iris",
  ui = function(id) {
    ns <- NS(id)
    tags$div(
      numericInput(ns("n_rows"), "Number of rows to display", value = 6, min = 1, max = 150, step = 1)
    )
  },
  server = function(id, data) {
    moduleServer(id, function(input, output, session) {
      reactive({
        within(
          data(),
          iris <- head(iris, num_rows),
          num_rows = input$n_rows
        )
      })
    })
  }
)

app <- init(
  data = data,
  modules = example_module(transformators = list(transformator_iris))
)

if (interactive()) {
  shinyApp(app$ui, app$server)
}

Note: The server function of a teal_transform_module must return a reactive expression with a teal_data object. In order to maintain full reactivity, we recommended using reactive() over eventReactive(). If you do use eventReactive() or bindEvent(), the trigger event should include data() (e.g. eventReactive(list(input$a, data()), {...})). See this discussion for a detailed explanation.

Multiple transformators

module(transformators) accepts a list, so we can use multiple teal_transform_modules at the same time.

Targeting multiple datasets

Let us add another transformation that creates a column with rownames in mtcars. Note that this module does not have interactive UI elements.

data <- within(teal_data(), {
  iris <- iris
  mtcars <- mtcars
})

transformator_iris <- teal_transform_module(
  label = "Custom transformator for iris",
  ui = function(id) {
    ns <- NS(id)
    tags$div(
      numericInput(ns("n_rows"), "Number of rows to subset", value = 6, min = 1, max = 150, step = 1)
    )
  },
  server = function(id, data) {
    moduleServer(id, function(input, output, session) {
      reactive({
        within(
          data(),
          iris <- head(iris, num_rows),
          num_rows = input$n_rows
        )
      })
    })
  }
)

transformator_mtcars <- teal_transform_module(
  label = "Custom transformator for mtcars",
  ui = function(id) {
    ns <- NS(id)
    tags$div(
      "Adding rownames column to mtcars"
    )
  },
  server = function(id, data) {
    moduleServer(id, function(input, output, session) {
      reactive({
        within(data(), {
          mtcars$rownames <- rownames(mtcars)
          rownames(mtcars) <- NULL
        })
      })
    })
  }
)

my_transformators <- list(
  transformator_iris,
  transformator_mtcars
)

app <- init(
  data = data,
  modules = example_module(transformators = my_transformators)
)

if (interactive()) {
  shinyApp(app$ui, app$server)
}

Targeting one dataset

It is also possible to have multiple transformator modules act on one dataset. In such cases, transformations will be executed in the same order in which the transformator modules are passed to the module.

data <- within(teal_data(), {
  iris <- iris
  mtcars <- mtcars
})

transformator_iris_scale <- teal_transform_module(
  label = "Scaling transformator for iris",
  ui = function(id) {
    ns <- NS(id)
    uiOutput(ns("scaled_columns_container"))
  },
  server = function(id, data) {
    moduleServer(id, function(input, output, session) {
      ns <- session$ns

      scalable_columns <- names(Filter(is.numeric, data()[["iris"]])) |> isolate()

      output$scaled_columns_container <- renderUI({
        selectInput(
          inputId = ns("scaled_columns"),
          label = "Columns to scale",
          choices = scalable_columns,
          selected = input$scaled_columns,
          multiple = TRUE
        )
      })

      reactive({
        within(
          data(),
          {
            iris[scaled_columns] <- scale(iris[scaled_columns])
          },
          scaled_columns = input$scaled_columns
        )
      })
    })
  }
)

transformator_iris <- teal_transform_module(
  label = "Custom transformator for iris",
  ui = function(id) {
    ns <- NS(id)
    tags$div(
      numericInput(ns("n_rows"), "Number of rows to subset", value = 6, min = 1, max = 150, step = 1)
    )
  },
  server = function(id, data) {
    moduleServer(id, function(input, output, session) {
      reactive({
        within(
          data(),
          iris <- head(iris, num_rows),
          num_rows = input$n_rows
        )
      })
    })
  }
)

transformator_mtcars <- teal_transform_module(
  label = "Custom transformator for mtcars",
  ui = function(id) {
    ns <- NS(id)
    tags$div(
      "Adding rownames column to mtcars"
    )
  },
  server = function(id, data) {
    moduleServer(id, function(input, output, session) {
      reactive({
        within(data(), {
          mtcars$rownames <- rownames(mtcars)
          rownames(mtcars) <- NULL
        })
      })
    })
  }
)

my_transformators <- list(
  transformator_iris,
  transformator_iris_scale,
  transformator_mtcars
)

app <- init(
  data = data,
  modules = example_module(transformators = my_transformators)
)

if (interactive()) {
  shinyApp(app$ui, app$server)
}

This approach provides greater flexibility in data preprocessing, allowing transformations to be tailored to specific datasets for a specific module.