An opinionated lightweight template for smooth drake flows.
# remotes::install_github("milesmcbain/dflow")
install.packages("dflow", repos = c('https://milesmcbain.r-universe.dev', 'https://cloud.r-project.org'))
Set dependencies = TRUE to also install capsule, conflicted, dontenv, and drake.
dflow::use_dflow():
./
|_ R/
| |_ plan.R
|
|_ _drake.R
|_ packages.R
|_ .env
dflow::use_rmd("analysis.Rmd"):
v Creating 'doc/'
v Writing 'doc/analysis.Rmd'
Add this target to your drake plan:
target_name = target(
command = {
rmarkdown::render(knitr_in("doc/analysis.Rmd")),
file_out("doc/analysis.html")
}
)
(change output extension as appropriate if output is not html)
library(rmarkdown) added to ./packages.R
dflow::use_gitignore():
Drop in a starter ./.gitignore with ignores for drake and renv among others.
dflow tries to set up a minimalist ergonomic workflow for drake pipeline
development. To get the most out of it follow these tips:
Put all your target code in separate functions in R/. Use fnmate to
quickly generate function definitions in the right place. Let plan.R define
the structure of the workflow and use it as a map for your sources. Use 'jump
to function' to quickly navigate to them.
Use a call r_make() to kick off building your plan in a new R session (via
callr). _drake.R is setup to make this work. Bind a keyboard shortcut to
this using the addin in drake.
Put all your library() calls into packages.R. This way you'll have them
in one place when you go to add sandboxing with renv, packarat, and
switchr etc.
Take advantage of automation for loading drake targets at the cursor with the 'loadd target at cursor' addin.
Some things are baked into the template that will help you avoid common pitfalls and make your project more reproducible:
library(conflicted) is called in packages.R to detect package masking issues.
.env is added carrying the following options to avoid misuse of logical vector tests:
_R_CHECK_LENGTH_1_LOGIC2_=verbose
_R_CHECK_LENGTH_1_CONDITION_=true