README.md

In jameelalsalam/analysispkg: Example Analysis Structured as a Package

analysispkg

The goal of analysispkg is to work out a structure for complex analyses-as-package.

Installation

Two ways to get the code:

If you just want to browse the or use the final results in another project, you can install the development version of analysispkg from Github with:

remotes::install_github("jameelalsalam/analysispkg")

If you want to re-run or modify the analysis, you must or download or clone the repository.

Running the Analysis

To re-run the analysis, open the analysispkg project. The package functions and dependencies must be loaded.

To ensure you have all dependencies:

devtools::install_deps(".")

To load package functions:

devtools::load_all(".")

And re-run the analysis plan using:

make(plan)

Goals & Features

The analysispkg structure has the following high level goals:

1) Facilitate re-use of analysis results by other projects,

2) Support iterative change and improvement during development.

3) Reproducible and transparent analysis results,

Features that support these goals:

Re-usability: * Final results are exported as datasets so they can be used in other projects by installing and loading the package. This is accomplished by including them as binary data in /data.

• Major analysis steps are implemented as functions. Function documentation serves as a methodological reference for these major steps.

Iterative Development: * The entire analysis is run with make.R and can be easily re-run during development (or by cloning the repository) using {drake} for caching and inter-dependency management.

• Analysis is re-run in C.I. on Github Actions, demonstrating that all resources are available and that changes in the code haven't broken the analysis.

Reproducibility and transparency: * Input data are stored in /data-raw or acquired externally

• Results are displayed through vignettes/articles (???)

Outstanding Questions:

• How to ensure there aren't conflicts between package data (which is the result of the analysis) and steps in the analysis. Idea 1: data exports are outputs of the drake plan, but with different names. Idea 2: a version of load_all which only loads code. Idea 3: non-package Miles McBain approach of sourcing all R files.

• Where should input data to the analysis be kept? Should it be in the repo at all? Should it be in the installed package?

• Should the analysis be built in C.I. or locally only? Should report results be built in C.I.?

• Is there a way to store the binary data as an artifact? I'd like to gitignore it.

• Drake can be used in multiple ways. Is _drake.R necessary? Should the drake plan be exported? Are some stuff in /inst or no?

How to Get Here:

1) usethis::create_package("yourpkgname") # from containing folder (or if you are already going, usethis::use_description() to pkg-ify)

2) Verify using roxygen2 to document

3) Add packages, e.g., usethis::use_package

4) Add drake infrastructure

5) Document package, exported data, and functions usethis::use_package_doc()

6) Add Github Actions basic infrastructure Modify it to be appropriate to analysis pkg use.

jameelalsalam/analysispkg documentation built on June 18, 2020, 12:15 a.m.