In baumer-lab/fertile: Creating Optimal Conditions for Reproducibility

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(readr)
library(fertile)

fertile: creating optimal conditions for reproducibility

Goal

• Make creating a reproducible project as easy as possible, for users of all levels of sophistication

Why do we need this?

• Creating an R package ensures some degree of quality control in the code, but is way too hard to for all but the most experienced R users. It's developer-centric.
• Most people, even experienced R users who value reproducibility, are terrible at actually creating reproducible data analysis projects
• Learning make is hard and maybe unnecessary
• Many journals now have reproducibility editors, but there is no standard or automated way to achieve reproducibility

What is it?

• goodpractice/R CMD check for data analysis projects

Prospective Reproducibility

"I'm fiddling with pineapple coconut cake recipes,... but I want someone else to be able to follow my steps."

Logs your actions

log_clear()
library(fs)

read_csv("data.csv")
read_csv("../tests/testthat/data/data.csv")
log_report()

Checks paths

readr::read_csv("../fertile/tests/testthat/data/data.csv")
fertile::read_csv("../fertile/tests/testthat/data/data.csv")
read_csv(tempfile())

write_csv(data.frame(a = 1), file = tempfile())
write_csv(data.frame(a = 1), file = "temp_data.csv")
log_report()

Prevents you from doing bad things

setwd("vignettes")

Works by shimming many I/O functions

?shims

Retrospective Reproducibility

"I've baked a pineapple cococnut cake. What was my recipe?"

• analyzes your code
• analyzes your files
• makes suggestions
• make moves

A simple project

unzip(system.file("extdata/project_noob.zip", package = "fertile"), exdir = "../tests/testthat")

project_noob <- here::here("tests", "testthat", "project_noob")
fs::dir_info(project_noob)

Test it

proj_test(project_noob)

A slightly more complicated project

unzip(system.file("extdata/project_miceps.zip", package = "fertile"), exdir = "../tests/testthat")

project_miceps <- here::here("tests", "testthat", "project_miceps")
fs::dir_info(project_miceps)

x <- proj_analyze(project_miceps)
x

proj_suggest_moves(x$suggestions)
proj_move_files(x$suggestions)
fs::dir_info(project_miceps)

fs::dir_delete(project_noob)
fs::dir_delete(project_miceps)

Moving forward

• auto-detect unnecessary package loads
• R CMD check style checklist of potential problems
• Incorporate some of kind of make-inspired build chain
• Unfakeable certificate of reproducibility?

Questions

• Will this actually work?? Will it be useful for anyone?
• What recommendations to make about file structure?
• How strict/opinionated to be?

baumer-lab/fertile documentation built on April 19, 2021, 12:01 a.m.