The Stan models live in
code/
(e.g.,
code/bsplines.stan
,
code/csplines.stan
) with companion R helpers such as
code/smoothing_diagnostics.R
. High level examples reside in
examples/
, including combined Stan/R workflows like
examples/hierarchical_regional_splines.R
. Tests are plain R scripts under
tests/
, each prefixed with
test_
and targeting a single behavior. Generated charts, CSVs, and CmdStan outputs should land in
output/
. Root-level runners (
run-code.R
,
run-examples.R
,
run-tests.R
) orchestrate minimal demos, extended analyses, and the full regression suite.
Rscript run-code.R
: Installs any missing R packages and runs the minimal spline demos, writing plots to output/
.Rscript run-examples.R
: Executes the comparative and hierarchical examples; expect longer runtimes because each Stan model compiles.Rscript run-tests.R
: Sources every tests/test_*.R
file, orders quick checks first, and reports per-script timing.cmdstanr::cmdstan_model("code/bsplines.stan")
: Use within R to compile a single model when iterating on Stan code.
Use two-space indentation for Stan blocks and keep identifiers in
snake_case
(
build_b_spline
,
smoothing_strength
). Comment intent with
//
before complex blocks rather than inline fragments. R scripts prefer tidyverse conventions:
<-
for assignment,
snake_case
filenames, and explicit
library()
calls in setup sections. Store reusable helpers beside their Stan counterparts in
code/
, and keep test utilities local to the test script that owns them.
All tests rely on
groundhog.library()
to lock package versions; add new dependencies to the vectors at the top of each test. Follow the naming format
tests/test_topic.R
and structure work as executable scripts that call CmdStan and assert diagnostics. Before opening a PR, run
Rscript run-tests.R
and confirm any new artifact paths point to
output/
. When adding Stan features, include at least one deterministic check (e.g., known function fit) to guard against regression.
Recent commits use concise, sentence-case summaries starting with a verb (
Improve hierarchical spline model ...
). Keep a single focus per commit and note material Stan or R dependency changes in the body. PRs should include: a short description of the behavior change, links to related issues, confirmation that
run-tests.R
completed, and key plot snapshots or summary tables when output shifts. Request at least one reviewer familiar with both Stan and R when modifying shared helpers.