knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
options(tibble.print_min = 3, tibble.print_max = 3)
delete_on_exit <- character()

Ideas for using readxl to increase reproducibility and reduce tedium.

Reproducibility is much easier in theory than in reality. Here are some special dilemmas we face with spreadsheets:

The examples below also demonstrate the use of functional programming or "apply" techniques to iterate over the worksheets in a workbook.

Load packages

We load the tidyverse metapackage here because the workflows below show readxl working with readr, purrr, etc. See the last section for solutions using base R only (other than readxl).

We must load readxl explicitly because it is not part of the core tidyverse.

library(tidyverse)
library(readxl)

Cache a CSV snapshot

Break analyses into logical steps, via a series of scripts that relate to one theme, such as "clean the data" or "make exploratory and diagnostic plots".

This forces you to transmit info from step i to step i + 1 via a set of output files. The cumulative outputs of steps 1, 2, ..., i are available as inputs for steps i + 1 and beyond.

These outputs constitute an API for your analysis, i.e. they provide clean entry points that can be used (and understood) in isolation, possibly using an entirely different toolkit. Contrast this with the alternative of writing one monolithic script or transmitting entire workspaces via save(), load(), and R-specific .rds files.

If raw data is stored only as an Excel spreadsheet, this limits your ability to inspect it when solving the little puzzles that crop up in dowstream work. You'll need to fire up Excel (or similar) and get busy with your mouse. You certainly can't poke around it or view diffs on GitHub.

Solution: cache a CSV snapshot of your raw data tables at the time of export. Even if you use read_excel() for end-to-end reproducibility, this complementary CSV leaves your analysis in a more accessible state.

Pipe the output of read_excel() directly into readr::write_csv() like so:

mtcars_xl <- readxl_example("datasets.xlsx") %>% 
  read_excel(sheet = "mtcars") %>% 
  write_csv("mtcars-raw.csv")
delete_on_exit <- c(delete_on_exit, "mtcars-raw.csv")

Why does this work? readr::write_csv() is a well-mannered "write" function: it does its main job and returns its input invisibly. The above command reads the iris sheet from readxl's datasets.xlsx example workbook and caches a CSV version of the resulting data frame to file.

Let's check. Did we still import the data? Did we write the CSV file?

mtcars_xl
dir(pattern = "mtcars")

Yes! Is the data written to CSV an exact copy of what we imported from Excel?

mtcars_alt <- read_csv("mtcars-raw.csv")
## readr leaves a note-to-self in `spec` that records its column guessing,
## so we remove that attribute before the check
attr(mtcars_alt, "spec") <- NULL
identical(mtcars_xl, mtcars_alt)

Yes! If we needed to restart or troubleshoot this fictional analysis, iris-raw.csv is available as a second, highly accessible alternative to datasets.xlsx.

Iterate over multiple worksheets in a workbook

Some Excel workbooks contain only data and you are tempted to ask "Why, God, why is this data stored in Excel? Why not store this as a series of CSV files?" One possible answer is this: because the workbook structure keeps them all together.

Let's accept that this happens and that it is not entirely crazy. How can you efficiently load all of that into R?

Here's how to load all the sheets in a workbook into a list of data frames:

path <- readxl_example("datasets.xlsx")
path %>% 
  excel_sheets() %>% 
  set_names() %>% 
  map(read_excel, path = path)

CSV caching and iterating over sheets

What if we want to read all the sheets in at once and simultaneously cache to CSV? We define read_then_csv() as read_excel(...) %>% write_csv() and use purrr::map() again.

read_then_csv <- function(sheet, path) {
  pathbase <- path %>%
    basename() %>%
    tools::file_path_sans_ext()
  path %>%
    read_excel(sheet = sheet) %>% 
    write_csv(paste0(pathbase, "-", sheet, ".csv"))
}

We could even define this on-the-fly as an anonymous function inside map(), but I think this is more readable.

path <- readxl_example("datasets.xlsx")
path %>%
  excel_sheets() %>%
  set_names() %>% 
  map(read_then_csv, path = path)
dir(pattern = "^datasets.*\\.csv$")
delete_on_exit <- c(delete_on_exit, list.files(pattern = "^datasets.*\\.csv$"))

In a real analysis, starting with workbook "foo.xlsx", you might want to create the directory foo and place the CSVs inside that.

Concatenate worksheets into one data frame

What if the datasets found on different sheets have the same variables? Then you'll want to row-bind them, after import, to form one big, beautiful data frame.

readxl ships with an example sheet deaths.xlsx, containing data on famous people who died in 2016 or 2017. It has two worksheets named "arts" and "other", but the spreadsheet layout is the same in each and the data tables have the same variables, e.g., name and date of death.

The map() function from purrr makes it easy to iterate over worksheets. Use purrr::list_rbind() to glue together the resulting data frames.

path <- readxl_example("deaths.xlsx")
deaths <- path %>%
  excel_sheets() %>%
  set_names() %>% 
  map(~ read_excel(path = path, sheet = .x, range = "A5:F15")) %>%
  list_rbind(names_to = "sheet")
print(deaths, n = Inf)

Note the use of range = "A5:E15" here. deaths.xlsx is a typical spreadsheet and includes a few non-data lines at the top and bottom and this argument specifies where the data rectangle lives.

Putting it all together

All at once now:

Even though the worksheets in deaths.xlsx have the same layout, we'll pretend they don't and specify the target rectangle in two different ways here. This shows how this can work if each worksheet has it's own peculiar geometry. Here's the workflow:

path <- readxl_example("deaths.xlsx")
sheets <- path %>%
  excel_sheets() %>% 
  set_names()
ranges <- list("A5:F15", cell_rows(5:15))
deaths <- map2(
  sheets,
  ranges,
  ~ read_excel(path, sheet = .x, range = .y)
) %>%
  list_rbind(names_to = "sheet") %>%
  write_csv("deaths.csv")
print(deaths, n = Inf)
delete_on_exit <- c(delete_on_exit, "deaths.csv")

Base version

Rework examples from above but using base R only, other than readxl.

Cache a CSV snapshot

mtcars_xl <- read_excel(readxl_example("datasets.xlsx"), sheet = "mtcars")
write.csv(iris_xl, "mtcars-raw.csv", row.names = FALSE, quote = FALSE)
mtcars_alt <- read.csv("mtcars-raw.csv", stringsAsFactors = FALSE)
## coerce iris_xl back to a data.frame
identical(as.data.frame(mtcars_xl), mtcars_alt)

Iterate over multiple worksheets in a workbook

path <- readxl_example("datasets.xls")
sheets <- excel_sheets(path)
xl_list <- lapply(excel_sheets(path), read_excel, path = path) 
names(xl_list) <- sheets

CSV caching and iterating over sheets

read_then_csv <- function(sheet, path) {
  pathbase <- tools::file_path_sans_ext(basename(path))
  df <- read_excel(path = path, sheet = sheet)
  write.csv(df, paste0(pathbase, "-", sheet, ".csv"),
            quote = FALSE, row.names = FALSE)
  df
}
path <- readxl_example("datasets.xlsx")
sheets <- excel_sheets(path)
xl_list <- lapply(excel_sheets(path), read_then_csv, path = path)
names(xl_list) <- sheets

Concatenate worksheets into one data frame

path <- readxl_example("deaths.xlsx")
sheets <- excel_sheets(path)
xl_list <-
  lapply(excel_sheets(path), read_excel, path = path, range = "A5:F15")
xl_list <- lapply(seq_along(sheets), function(i) {
  data.frame(sheet = I(sheets[i]), xl_list[[i]])
})
xl_list <- do.call(rbind, xl_list)

Putting it all together

path <- readxl_example("deaths.xlsx")
sheets <- excel_sheets(path)
ranges <- list("A5:F15", cell_rows(5:15))
xl_list <- mapply(function(x, y) {
  read_excel(path = path, sheet = x, range = y)
}, sheets, ranges, SIMPLIFY = FALSE)
xl_list <- lapply(seq_along(sheets), function(i) {
  data.frame(sheet = I(sheets[i]), xl_list[[i]])
})
xl_list <- do.call(rbind, xl_list)
write.csv(xl_list, "deaths.csv", row.names = FALSE, quote = FALSE)
delete_on_exit <- unique(delete_on_exit)
file.remove(delete_on_exit)


hadley/readxl documentation built on Aug. 25, 2024, 4:18 p.m.