jrozra200 / reproducibleProjects

Reproducible Projects

This package is designed to allow for and facilitate reproducible projects to be created in RStudio. It is really easy to work on a full analysis, deliver some output, then shelve the code; when you are asked to revisit it for whatever reason down the line, it can be a fraught exercise to figure out what you did.

To ensure that analysis projects are built from the outset in a reproducible way, this package was created.

The inspiration for this package comes from DrivenData's Cookiecutter Data Science Project and Rich Fitzjohn's Designing Projects Blog Post.

Also included in this template is DrivenData's deon Ethics Checklist.

Installation

This package will be submitted to cran - and these instructions will be updated if/when this package is available there. At this time, you can install directly from github.

devtools::install_github("jrozra200/reproducibleProjects")

After installation, you can create a new project using the reproducible project template: File > New Project > New Directory > Reproducible Project Template.

The Structure and Intention

There are four top level directories and one final RMarkdown document created at the outset. The full topology is:

Project/
|---- R/
|---- data/
      |---- external/
      |---- raw/
      |---- interim/
      |---- processed
|---- visualization/
|---- output/
|---- final_analysis.Rmd

R Directory

This is where all R code goes. It is best practice, though I am not sure I follow it, to functionalize all of your analysis, then call those functions in the final_analysis.Rmd document.

It would also be appropriate to include exploratory and other code that may not be ultimately used in the final output.

data Directory

There are 4 subdirectories here: external, raw, interim, and processed.

The raw data should never be changed (some might use the big word immutable). Keep that initial data here. Once you've edited it in anyway, it should then go in the interim (if it is not in the final state) or the processed (if it is finalized) directories.

If your project will ultimately ship some sort of data, it should be kept in the processed directory.

The external directory is meant to house any tangential datasets from external sources.

visualization Directory

This is where any data visualizations should be kept. Often the point of analysis is to create compelling data viz that will be copy/pasted into another medium (yep... thinking about you, PowerPoint). Your data viz should be properly labeled and stored here.

output Directory

Any other artifacts that arise as part of your analysis: logs, simulation, etc. should be placed here.

final_analysis.Rmd

This file should be the source of truth. When you come back to your analysis in 6 months, click knit and you should get the exact output of your analysis. This script will call the R functions, datasets, and then create viz and other outputs.

The Ethics Checklist

One important note: in the final_analysis.Rmd, the last section is DrivenData's deon Ethics Checklist. Not to dive into a rant, but ethics is the foundation to all things, data science especially. To ensure that it remains at the crux of all the things, I've included this checklist. By having it as part of your output, you have to actively remove it in order to be unethical.

What you do is important. Make sure it is ethical.

What is left to do?

This package needs to be submitted to cran. It also has not been used in the wild, yet, so it may need tweaks as I go.