Construct a linear basis feature space

A basis_space is a representational feature axis linked to one parent spatial space. Its encoder maps parent features to component coefficients; its optional decoder maps coefficients back to the parent. Fitting and model-specific offsets remain the responsibility of packages such as fmrilatent.

Usage

basis_space(
  parent,
  component_ids,
  encoder,
  decoder = NULL,
  data = NULL,
  basis_type = "linear_basis",
  provenance = list(),
  tolerance = 1e-08,
  metadata = list()
)

Arguments

parent

Parent feature_space represented by the basis.

component_ids

Stable component identifiers.

encoder

Component-by-parent analysis operator.

decoder

Optional parent-by-component synthesis operator.

data

One metadata row per component.

basis_type

Stable basis-family label.

provenance

Serializable derivation metadata.

tolerance

Maximum absolute error permitted when validating that encoder %*% decoder is the component-space identity.

metadata

Additional serializable metadata.

Value

A basis_space.

Examples

parent <- volume_space(c(2, 1, 1), support = 1:2)
x <- basis_space(parent, c("c1", "c2"), diag(2), diag(2))
n_features(x)
#> [1] 2