Contributions are welcome. The package is at an early stage, so the most useful contributions right now are the ones that test the design against real problems.
Ways to help that are genuinely useful
Try it on your data and report what broke. The package is currently validated on one simulated system and one real data set. Every new data set is a chance to find a wrong assumption. Open an issue with what you tried and what happened.
Add a data source. Providers are a function plus one registration call, and need no change to the package:
register_source("my_provider",
fetch = function(units, seasons, ...) {
# series: one row per unit_id, season_id and date
# static: one row per unit_id
},
provides = c("tmin", "tmax"),
kind = "series")Add a learner. Same pattern with register_learner(). Please wrap an existing, maintained implementation rather than writing a new one.
Calibrate crop parameters. The thermal thresholds in crop_parameters() are indicative defaults. Locally calibrated values, with a citation, are worth more than any amount of new code.
What is out of scope
Reimplementations of things CRAN already does well: learners, cross-validation algorithms, raster IO, satellite access. This package is an integration layer and delegates on purpose. See ARCHITECTURE.md for the reasoning.
Practical matters
- Every change needs a test. Tests must pass offline and must not require network access; gate anything that does with
skip_if_not_installed()orskip_on_cran(). -
R CMD checkmust stay clean, with no new warnings or notes. - Keep
Importsfree of heavy dependencies. Optional packages belong inSuggestsbehindrequireNamespace(). - Style: four-space indents, lines under 80 characters,
snake_case. - Run before opening a pull request:
devtools::document()
devtools::test()
devtools::check()Reporting a bug
A reproducible example matters more than a description. If the data are not shareable, demo_agri_data() generates a deterministic data set that often reproduces structural problems just as well.