Introduction"
In R6Frame: R6 Wrapper for Data Frames

R6Frames

R6Frame provides a R6 "frame" around a data.frame (or data.table), which allows one to create more complex objects/operations based on the underlying data. R6Frame is meant to be used as a template for R6 subclasses, and comes with S3 methods for most data.frame operations included.

As an example, R6Frames can be used to handle labels for variables in a data.frame. This allows us to preserve the labels more easily than using a label attribute for each variable.

Creating a subclass

To subclass R6Frame, we can simply specify it under inherit. In the code below, we create a minimal subclass with a private field called .label and public getter/setter methods for the field. Because we want to be able to for instance rbind and preserve labels, we also need a merge_vectors function which prioritizes the label for the first R6Frame in e.g. rbind:

# Subclass "Labelled_df" for R6Frame.
library(R6Frame)
is.labelled_df <- function(x) inherits(x, "Labelled_df")
Labelled_df <- R6::R6Class("Labelled_df",
  inherit = R6Frame::R6Frame,

  # Private methods
  private = list(.label = NULL),

  # Public methods
  public = list(
    set_label = function(..., list = NULL) {
      # Set label for a variable.
      new <- merge_vectors(..., list, private$.label, default = self$names())
      private$.label <- new
      invisible(self)
    },
    get_label = function(which = NULL) {
      # Get label for a specified variable. (NULL = all labels.)
      res <- private$.label
      if (!is.null(res) && !is.null(which)) {
        res <- res[match_all(which, names(res))]
        if (!length(res)) res <- NULL
      }
      res
    }
  )

)

# Utility function that merges named vectors for private field (.label) in Labelled_df.
# Duplicates are dropped from the end of the named vector (after unlisting).
merge_vectors <- function(..., default = NULL) {
  dots <- list(...)
  if (!length(dots)) stop("No vectors supplied.")

  # Use unnamed default's as names for a vector of NA's.
  # (So we can pass colummnames as a default.)
  if (!is.null(default) && is.null(names(default))) {
    default <- setNames(rep(NA, length(default)), default)
  }

  res <- c(unlist(dots), default)
  nms <- names(res)
  if (is.null(nms) && any(is.na(nms)) && any(nms == ""))
    stop("All elements must be named.")

  # Return should be ordered by default if it exists
  res <- res[!duplicated(names(res), fromLast = FALSE)]
  if (!is.null(default)) {
    res[names(default)]
  } else {
    res
  }

}

`R6Frame$do()`

Each function call on a R6Frame is sent to either $do() or $do_merge() (all binds and joins), which can be modified to update the .label field everytime we perform an operation on the data. For this example, we need to update $do_merge() for our labels to follow if we merge or join two or more Labelled_df:

Labelled_df$set(
  "public",
  "do_merge",
  function(f, dots, env) {
      # Get existing labels from data
      lab <- lapply(dots, function(x) { if (is.labelled_df(x)) x$get_label() })
      # Assign to private field (self$do and in turn self$update will remove duplicates)
      private$.label <- merge_vectors(private$.label, lab)
      # Call the default R6Frame method "do_merge" to complete the operation.
      super$do_merge(f, dots, env)
  }
)

`R6Frame$update()`

Whenever the result of an operation is a new (or the same) data.frame, R6Frame will call $update() on the result. In order to drop labels as variables are removed for instance, or add a "slot" for a new variable, we have to change $update() as follows:

Labelled_df$set(
  "public",
  "update",
  function(renamed = NULL) {
      # If renamed is not NULL, it is a vector with the same length as the data
      # and contains the new names for the variables. (Usually from dplyr methods.)
      if (!is.null(renamed)) {
        private$.label <- setNames(private$.label, renamed)
      }
      self$set_label()
      self
  }
)

Example usage

In the code below, we can see usage examples of our new Labelled_df class and how the labels are kept updated through various function calls.

org <- data.frame("A" = c("Yes","No"), "B" = c(1, 2), stringsAsFactors = FALSE)

# Initialize
df <- Labelled_df$new(org)

# Set label
df$set_label(A = "Yes/No", B = "Numbers")
df$get_label()

# Add a column
df[, "test"] <- "test"

# Split and set label
df2 <- df[, "test", drop = FALSE]
df[["test"]] <- NULL

df2$set_label(test = "Test label")

# Join again
df3 <- cbind(df2, df)
df3$get_label()