Themed extensions for workflowr research workflows
reflowR extends the workflowr package with themed colour schemes, structured analysis pipeline templates, and utility scripts for reproducible research. It wraps workflowr::wflow_start() to create a standard workflowr project, then layers on a complete theming and scaffolding system.
code/ -- setup, utils, formulary, data I/O, table helpers, plot helpersAll generated projects remain fully compatible with standard workflowr commands (wflow_build(), wflow_publish(), etc.).
reflowR is not on CRAN yet. Install it from GitHub:
if (!requireNamespace("pak", quietly = TRUE)) install.packages("pak")
# pak::pak("CTTIR/reflowR")
install.packages("reflowR", repos = c('https://cttir.r-universe.dev', 'https://cloud.r-project.org'))
library(reflowR)
# The generated setup requires here in addition to reflowR.
if (!requireNamespace("here", quietly = TRUE)) install.packages("here")
# Choose a new project directory.
dir.create("~/projects", recursive = TRUE, showWarnings = FALSE)
project <- reflow_init(
directory = "~/projects/my_analysis",
author = "Your Name",
email = "you@example.com",
scheme = "clinical",
depth = "standard",
change_wd = FALSE,
open = FALSE
)
# Then use standard workflowr commands:
workflowr::wflow_build(project = project)
workflowr::wflow_publish(file.path(project, "analysis", "*.Rmd"),
message = "Initial build",
project = project)
The explicit project argument and project-qualified analysis paths work from
your current directory; initialization does not need to change it. Choose a new destination for this example. Generated
code/00_setup.R reports any additional packages needed by individual analysis
helpers. Install those packages before using the corresponding helpers; the
scaffold does not install them automatically. Publishing with workflowr commits
the reviewed analysis and rendered files locally; remote hosting is a separate
setup step. The optional table helpers use gtsummary >= 2.0.0, which
requires R >= 4.2; core reflowR remains available on R >= 4.1.
Generated save_versioned() checkpoints retain their second-resolution names.
A same-name, same-second collision is refused and the earlier checkpoint is
preserved. Loading matches the exact step name, including punctuation, then
selects the latest modification time among matching timestamped files.
Serialization completes in a temporary file before publication, so a failed
write does not leave a partial named checkpoint. This is not a guarantee
for concurrent writers or power-loss recovery. numeric_summary() reports
n = 0 with NA estimates for empty or entirely missing numeric columns.
| Scheme | Label | Accent | Use Case |
|---|---|---|---|
spectral |
Spectral imaging | #0072B2 |
Imaging, spectroscopy |
spatial |
Spatial biology | #009E73 |
Tissue and spatial assays |
omics |
Molecular data | #CC79A7 |
Sequencing, molecular profiles |
clinical |
Clinical science | #D55E00 |
Clinical trials, patient data |
methods |
Learning & inference | #56B4E9 |
Statistical and ML methods |
simulation |
Systems & simulation | #E69F00 |
Simulation studies, models |
evidence |
Evidence & discovery | #F0E442 |
Reviews, meta-analyses |
platform |
Research infrastructure | #000000 |
Software, data, infrastructure |
brand |
Science, connected | #0072B2 |
General purpose (default) |
The earlier names basic, code, special and other remain accepted as
aliases of brand, methods, evidence and platform. Links and text use
neutral tokens in light and dark mode; the accent marks identity only.
Preview any scheme interactively:
reflow_preview("clinical")
reflow_schemes()
| Preset | Steps | Best For |
|---|---|---|
minimal |
3 | Quick explorations, side analyses |
standard |
8 | Standard research projects, publications |
extended |
12 | Clinical trials, comprehensive studies |
reflow_presets()
reflowR calls workflowr::wflow_start() to create the standard workflowr scaffold, then enhances it with themed files. The result is a standard workflowr project -- all workflowr commands work as expected.
citation("reflowR")
Contributions are welcome. Please open an issue first to discuss what you would like to change, then submit a pull request.
MIT
The imaging API runs a sequential graph of exported package functions, with explicit references to upstream R objects. It is independent of workflowr scaffolding and does not create Git repositories or publish anything.
plan <- reflow_imaging_plan(
reflow_imaging_stage("summary", "base", "summary", list(object = 1:4)),
reflow_imaging_stage("mean", "base", "mean",
list(x = reflow_imaging_ref("summary")))
)
run_dir <- tempfile("imaging-run-")
results <- reflow_imaging_run(plan, run_dir)
verified <- reflow_imaging_resume(plan, run_dir)
Use reflow_imaging_input(path) for tracked files or directories, or set
read = TRUE to pass an RDS input as an object. Use the reference's select
argument to extract named nested components. Declare optional package backends
with the plan's packages argument and stochastic calls with a stage seed.
Receipts record configuration, input, installed code, runtime and output hashes. Resume verifies completed results; changed inputs or corrupt results require a new run directory. Common numerical thread environment settings are recorded and changes invalidate resume; this does not measure effective backend thread counts or capture in-process thread setters. Failed or interrupted calls retry in new attempt directories. A lock prevents concurrent execution; stale locks require manual inspection. Inputs must remain immutable during execution. Calls must be trusted, return plain serializable R objects, and avoid external side effects: this is not a sandbox. Resource scheduling, file-producing stages, external engine provenance, review dashboards and report recipes are not yet implemented by this API.
Where applicable, LLM tools were used for narrowly defined, non-authorial tasks: copyediting, prose refinement, Markdown/LaTeX formatting, boilerplate scaffolding (such as CI configurations and build scripts), and code refactoring. Tools included Chat AI, the KISSKI (GWDG) service, and self-hosted Mistral Small (24B, Apache-2.0) served through a local vLLM architecture on an Ubuntu Linux workstation with Intel processors and NVIDIA GPUs. The self-hosted model ran locally, with no data sent to third parties.
All scientific claims, methodological choices, analyses, interpretations and conclusions remain the authors' own. Where applicable, LLM-generated material must be reviewed and revised before inclusion, and references must be verified against their DOI, arXiv ID or ISBN.
Local software setup: vLLM with Python/PyTorch, plus R and RStudio.
Editable R artwork and regeneration instructions are in tools/branding.