Extending hypermodR

Extensions must preserve identity, units, input semantics and explicit failures. This small local S3 bridge demonstrates conversion to the public S4 container; it is not a registration of a new qualified camera or model backend.

hm_as_dataset.example_spectra <- function(x, profile = NULL, ...) {
  hm_dataset(x$values, x$observations, x$bands,
    provenance = list(synthetic = TRUE, source = "local example")
  )
}
source <- structure(list(
  values = matrix(1:6, 2, 3),
  observations = data.frame(observation_id = c("a", "b")),
  bands = data.frame(wavelength = c(500, 600, 700))
), class = "example_spectra")
x <- hm_as_dataset(source)
if (!isTRUE(all(
  methods::validObject(x),
  identical(as.numeric(hm_spectra(x)$values), as.numeric(source$values))
))) {
  cli::cli_abort("Example assertion failed.", class = "hm_example_error")
}

A distributed adapter needs registered S3/S4 dispatch and conformance tests for identity, native axis order, explicit wavelength availability, units, calibration, geometry, masks, annotation revision and temporal order. Use asymmetric fixtures and independent expected values. Reject unsupported conversions before work.

There is no public arbitrary model-plugin registration API. New model engines must integrate the package’s task/specification contract with train-only recipe fitting, prediction schema and explicit fit/predict availability. Backend extensions require installed resources addressed through system.file(), exact artifact hashes/rights and numerical parity. Do not load code from mutable remote repositories or use private development paths at runtime. Read the architecture, interoperability and backend specifications for acceptance criteria.