R/summarize-vars.R

In ck37/ck37r: Chris Kennedy's R Toolkit

#' Summary table of variables
#'
#' Generate a summary table of variables in a dataframe.
#'
#' @param df Dataframe of variables
#' @param vars Variable names to analyze in the dataframe
#' @param few_uniq_vals Number of values below which we display the numeric variable
#'   as a candidate to be an integer.
#' @param groups TBD
#' @param integers Explicitly note integer-valued variables
#' @param ordinal Explicitly note ordinal variables
#'
#' @importFrom magrittr %>%
#' @export
summarize_vars =
  function(df,
           vars = names(df),
           few_uniq_vals = 10L,
           groups = NULL,
           integers = NULL,
           ordinal = NULL) {

  var_df = data.frame(matrix(nrow = length(vars),
                             ncol = 13L))
  names(var_df) = c("var", "group", "class", "type",
                    "uniq_vals", "mode", "mean", "median",
                    "min", "pctile_0.1", "max", "pctile_99.9", "missingness")

  var_df$var = vars
  var_df$class = sapply(df[, vars], class)
  var_df$mode = sapply(df[, vars], function(var) ck37r::Mode(var)[1])

  # Don't count NAs as a unique value.
  var_df$uniq_vals = sapply(df[, vars], function(var) length(setdiff(unique(var), NA)))

  summary_stats = lapply(vars, function(var_name, data) {
    var = data[[var_name]]
    summ = summary(var)

    # This will fail on string variables, and no need to do it for factors.
    # We use any() in case var has multiple classes, e.g. "labelled" and "factor".
    if (!(any(class(var) %in% c("factor", "character")))) {
      tryCatch({
        var_cdf = stats::ecdf(var)
        # TODO: set these as function arguments.
        summ["pctile_99.9"] = stats::quantile(var_cdf, 0.999)
        summ["pctile_0.1"] = stats::quantile(var_cdf, 0.001)
      },  error = function(e) {
        cat("Error while processing", var_name, "\n")
        print(e)
        summ["pctile_99.9"] = NA
        summ["pctile_0.1"] = NA

      })
    } else {
      # No need to report these quantiles for strings/factors.
      summ["pctile_99.9"] = NA
      summ["pctile_0.1"] = NA
    }

    summ
  }, df)

  var_df$mean = sapply(summary_stats, `[`, "Mean")
  var_df$min = sapply(summary_stats, `[`, "Min.")
  var_df$max = sapply(summary_stats, `[`, "Max.")
  var_df$median = sapply(summary_stats, `[`, "Median")

  # TODO: make these function arguments.
  var_df$pctile_99.9 = sapply(summary_stats, `[`, "pctile_99.9")
  var_df$pctile_0.1 = sapply(summary_stats, `[`, "pctile_0.1")

  var_df$missingness = colMeans(is.na(df[, vars]))

  # Put all variables into the same group.
  if (is.null(names(groups))) {
    groups = list("all" = vars)
  }

  for (group in names(groups)) {
    var_df$group[var_df$var %in% groups[[group]]] = group
  }

  if (is.null(integers)) {
    integers = names(which(sapply(df[, vars], is.integer)))
  }

  # Factor variables are category.
  var_df$type[var_df$class == "factor"] = "categorical"

  # Programatically identify binary vars.
  binary_vars = var_df$class %in% c("numeric", "integer", "boolean") &
    var_df$min == 0 & var_df$max == 1 & var_df$uniq_vals == 2

  # Recode binary vars to integers.
  for (binary_var in vars[binary_vars]) {
    df[[binary_var]] = as.integer(df[[binary_var]])
  }

  # Update type
  var_df$type[binary_vars] = "binary"

  for (integer_var in integers) {
    df[[integer_var]] = as.integer(df[[integer_var]])
  }

  # TODO: detect positive integers automatically.
  var_df$type[var_df$var %in% integers] = "integer"

  for (ordinal_var in ordinal) {
    df[[ordinal_var]] = as.integer(df[[ordinal_var]])
  }

  # These are all nonnegative integers
  var_df$type[var_df$var %in% ordinal] = "ordinal int."

  # Update class
  var_df$class = sapply(df[, vars], class)

  # Review numerics with few unique vals.
  few_unique_vals = var_df$class == "numeric" & var_df$uniq_vals <= few_uniq_vals

  for (var in var_df$var[few_unique_vals]) {
    cat("Var:", var, "\n")
    print(table(df[[var]], useNA = "ifany"))
  }

  # Make remaining numeric vars continuous type
  var_df$type[var_df$class == "numeric" & is.na(var_df$type)] = "continuous"

  # Convert group to a factor with the original ordering, so that we order
  # the results by the group ordering.
  var_df$group = factor(var_df$group, levels = names(groups))

  # Sort by group and then variable name.
  var_df = var_df %>% dplyr::arrange(group, var) %>% as.data.frame()

  result = list(data = df,
                table = var_df)

  result
}