Computational Project Structure

Introduction

• Learn how to organize a computational project
• Many benefits to good organization & documentation
  • Understand all project components
  • Work more efficiently
  • Easier for collaboration
  • Including with your future self
• Save a lot of time & frustration in long run, even if it’s time-consuming and frustrating to do it right initially

Example project walk-through

Download example project.

• Good file structure
  • All project files in one main folder
  • Subfolders (data, R, etc.)
• Main folder is R project
  • Self-contained project
  • Use relative instead of absolute paths
• Good folder & file names
  • Descriptive but not too long
  • No spaces
  • Consistent format
• README
  • Everything someone needs to know to understand project
• Raw data
  • In separate folder from cleaned data
  • Never change!
  • Each file should have metadata
• Scripts with code
  • Relative file paths to read in and create files
  • Lots of comments
  • Order: libraries, data, user-created functions, everything else
  • Good variable & column names
• Two methods to deal with different versions of same file
  • Good = consistent naming scheme
    • Each copy has date or version number
    • Don’t call anything final, cause no file ever is!
  • Better = version control, only have one copy of each file
• Running all scripts
  • Good = description of order to run each script in README
  • Better = script to run all scripts
  • Best = Make
• Combine code output and text in single document, e.g., knitr

Goal of good project structure

• Reproducibility = ability to get same results from set of data & code
• Give entire project folder to someone else (or your future self!), they can understand all parts and recreate your results, figures, documents, etc.
• Improving reproducibility makes YOUR life easier!