hypermodR brings hyperspectral data, train-only preprocessing, grouped modelling and explicit evaluation into one R workflow. It is development software for research. Executable interfaces, technical qualification and scientific validity are separate states. No clinical performance or general instrument compatibility is claimed.
| Route | Current scope | Qualification boundary |
|---|---|---|
| R models | Grouped statistical, chemometric and conventional learning adapters | Synthetic tests; optional R engines require their packages |
| Data bridges | S4 datasets/sequences, S3 source conversion and bounded blocks | Synthetic and companion fixtures; camera profiles remain unqualified |
| Neural models | Package-owned trainable references and named upstream admission contracts | No named pretrained model scientifically qualified |
| Runtime setup | Explicit resolver, reviewed manifest, isolated installation and owned workers | Automated resolver: Linux x86_64 only; exact live smoke/parity required |
| Workbench | Local Shiny workflow using the same package API | Browser/accessibility qualification tracked separately |
From a checked-out source directory:
install.packages(".", repos = NULL, type = "source")Install optional R engines only when needed. Loading hypermodR does not
install Python, download weights or start a server. The package is not
presented as a CRAN release. Read the installed runtime setup article
with vignette("backends-and-installation", package = "hypermodR").
The task is continuous response prediction from eight synthetic spectral bands (500–850 nm, arbitrary radiometric units). Subjects are the independent split units; the 96 spectra are repeated measurements from 12 subjects. This is one outer fold for illustration, not a selected model or confirmation study. Scaling is learned on training rows only. The held-out RMSE is reconstructed independently before an immutable run is saved and read back.
data <- hm_example(seed = 42L)
if (!isTRUE(all(
methods::validObject(data), nrow(data@observations) == 96L,
identical(data@bands$wavelength, seq(500, 850, length.out = 8L))
))) {
cli::cli_abort("Example assertion failed.", class = "hm_example_error")
}
task <- hm_task("response", "regression", "spectrum", "subject_id")
splits <- hm_split(data, task, folds = 3L, inner_folds = 2L, seed = 7L)
fold <- splits$folds[[1L]]
if (!isTRUE(all(!length(intersect(
data@observations$subject_id[fold$train],
data@observations$subject_id[fold$test]
))))) {
cli::cli_abort("Example assertion failed.", class = "hm_example_error")
}
workflow <- hm_workflow(
task, hm_recipe(list(list(type = "scale"))),
hm_model_spec("lm", task)
)
fit <- hm_fit(workflow, data, fold)
heldout <- hm_spectra(data, list(
observation_id = data@observations$observation_id[fold$test]
))
prediction <- hm_predict(fit, heldout)
evaluation <- hm_evaluate(prediction, heldout$observations)
errors <- prediction$values - heldout$observations$response
expected_rmse <- sqrt(mean(errors^2))
if (!isTRUE(all(isTRUE(all.equal(unname(evaluation$metrics["rmse"]),
expected_rmse))))) {
cli::cli_abort("Example assertion failed.", class = "hm_example_error")
}
evaluation
#> rmse mae bias
#> 0.10558376 0.08430486 0.03453047
uncertainty <- hm_bootstrap(evaluation, unit = "subject_id", draws = 20L,
seed = 9L)
if (!isTRUE(all(uncertainty$invalid_draws == 0L))) {
cli::cli_abort("Example assertion failed.", class = "hm_example_error")
}
uncertainty$conditioning
#> [1] "fixed predictions; fitting and selection uncertainty excluded"
run <- tempfile("hypermodR-run-")
hm_save_run(list(
fit = fit, prediction = prediction, evaluation = evaluation,
uncertainty = uncertainty
), run)
restored <- hm_read_run(run)
if (!isTRUE(all(identical(restored$prediction, prediction)))) {
cli::cli_abort("Example assertion failed.", class = "hm_example_error")
}
unlink(run, recursive = TRUE)HMDataset and HMSequence are formal S4 containers. hm_spectra()
and hm_read_block() are S4 accessors; hm_as_dataset() bridges
supported S3 sources. hm_fit and hm_prediction are S3 results with
standard print/summary and prediction methods. Camera brands do not
establish wavelength calibration. TIVITA profiles distinguish spectral
cubes, RGB and vendor displays; Cubert sequences retain frame identity,
explicit time and calibration declarations. See the article
catalogue.
hm_models() exposes model identities and support status.
hm_model_spec() is the common constructor. Named upstream constructors
are hm_htc(), hm_hyperfree(), hm_hypersigma(), hm_hypervision(),
hm_dual_stream_mhsi(), hm_spectralearth() and hm_spectralgpt(). A
registry entry is not an executable or qualified model.
hm_temporal_model() declares causal package reference families. Stock
vLLM serves compatible language/vision models; it does not execute these
native HSI architectures.
hm_app()The local app offers a synthetic example and reproducible exports
through the same task, workflow, split and run APIs. Read the installed
workbench guide with
vignette("shiny-workbench", package = "hypermodR").
Spectralverse is the central
external reference programme:
HeiPor anchors
primary anatomy,
xeno
species/perfusion development, and
ManU separate
human admission. hm_benchmark_protocol() validates protocol metadata;
it does not supply data, grant rights or certify external validation.
Target outcome access, case/ROI support and native versus adapted
spectral tracks remain explicit.
Bootstrap intervals condition on fixed predictions and omit training/selection uncertainty. Development, transfer and confirmation claims require distinct independent units and admission evidence. Missing-class draws remain invalid; model confidence is not an admitted out-of-distribution detector.
Package-owned code uses the MIT license. Data, external
source, SDKs and weights retain their own terms. Review the
contribution guide,
architecture
and quality
requirements.
Use citation("hypermodR") for the software citation; no publication
DOI is claimed.
Editable R sources and regeneration instructions are in tools/branding.