St. James Ballroom
Developing and Deploying Reproducible In-House R Packages: A Non-Packrat Workflow (303828)
*Eric Tesdahl, SpecialtyCare
Keywords: R, R package development, reproducibility
Relative to traditionally-used enterprise tools for data analysis, the R statistical computing environment with its framework of packages, affords a number of impressive capabilities for the creation of analytic tools that are transparent, well-documented, and easily-disseminated. However, R’s lack of an explicit package versioning mechanism puts the need for reproducibility of analytic findings in direct conflict with the continuous development and improvement of package functions. Packrat, the most commonly used tool to address reproducibility of R packages, unfortunately introduces significant resource and logistical constraints for most analytic workflows, and does not lend itself well to use by novice-level R users. This presentation describes one response to these challenges: a workflow for developing and distributing in-house R packages, which employs an explicit versioning system to ensure reproducible outputs, even as package code is updated over time.