Fit or apply the ROCKET transform

Uses aeon.transformations.collection.convolution_based.Rocket through an optional, caller-configured reticulate backend. With model = NULL, fits only on the selected cases and returns a reusable persistent model. With a model, validates the complete input contract and transforms without refitting.

Usage

rocket(
  x,
  model = NULL,
  assay_name = NULL,
  channels = NULL,
  cases = NULL,
  n_kernels = 10000L,
  seed = 1L,
  normalize = TRUE,
  backend = "aeon"
)

Arguments

x

A PhysioExperiment.

model

NULL or a fitted physio_rocket_model.

assay_name, channels, cases

Input selection; see catch22().

n_kernels

One finite whole number of kernels.

seed

One whole-number backend seed in [0, 2^31 - 1].

normalize

One non-missing logical passed to aeon's normalise.

backend

Exactly "aeon".

Value

A physio_rocket_transform containing a finite case-by-feature matrix, exact case identities, a persistent fitted model, settings, and typed diagnostics.

Details

Input assays are time x channel or time x channel x case and must contain finite, equal-length series of at least nine samples. The fitted model binds exact channel labels/order and time length. Requested kernel count and realized feature width are recorded separately because backend transforms may adjust the width.

Models persist the backend with Python pickle in a hashed raw vector. A matching SHA-256 detects corruption but does not make deserialization safe: only reuse model objects from trusted sources. PhysioML validates class, metadata, hash, input identity, and backend version before unpickling.

References

Dempster A, Petitjean F, Webb GI (2020). ROCKET: Exceptionally fast and accurate time series classification using random convolutional kernels. Data Mining and Knowledge Discovery, 34, 1454-1495. doi:10.1007/s10618-020-00701-z

Examples

arr <- array(stats::rnorm(60 * 2 * 4), dim = c(60, 2, 4))
pe <- PhysioExperiment::PhysioExperiment(
  assays = list(raw = arr),
  colData = S4Vectors::DataFrame(label = c("C3", "C4")),
  samplingRate = 100
)
# \donttest{
# The transform needs a caller-managed Python environment with NumPy and aeon.
fit <- rocket(pe, n_kernels = 50)
#> Error: the active reticulate Python lacks required module(s): aeon
dim(fit$features)
#> Error: object 'fit' not found
# }