setup_ICA — setup_ICA • rtemis

Setup ICA config.

Usage

setup_ICA(
  k = 3L,
  type = "parallel",
  fun = "logcosh",
  alpha = 1,
  row_norm = TRUE,
  maxit = 100L,
  tol = 1e-04,
  features = NULL
)

Arguments

k: Integer [1, Inf): Number of components.
type: Character {"parallel", "deflation"}: Type of ICA.
fun: Character {"logcosh", "exp"}: ICA function.
alpha: Numeric [1, 2]: Used in approximation to neg-entropy with fun = "logcosh".
row_norm: Logical: If TRUE, standardize each case across features before ICA, i.e. subtract each case's mean and divide by its standard deviation. Appropriate when the features share units and each case's overall level and scale are a nuisance rather than signal, as for spectra or assays whose total intensity varies by sample – the usual situation for the wide, single-instrument matrices decomposition is run on, which is why it is the default. It spends two degrees of freedom per case, so its cost is roughly 2 / ncol(x): negligible for hundreds of features, but substantial for a handful, where each case's level and scale are more likely to be signal. Set FALSE for narrow or mixed-unit data.
maxit: Integer [1, Inf): Maximum number of iterations.
tol: Numeric [0, Inf): Tolerance.
features: Optional Character vector: Names of at least 2 distinct feature columns to decompose. NULL decomposes all numeric features.

Value

ICAConfig object.

Author

EDG

Examples

ica_config <- setup_ICA(k = 3L)
ica_config

#> <ICA DecompositionConfig>
#>        k: <int> 3
#>     type: <chr> parallel
#>      fun: <chr> logcosh
#>    alpha: <nmr> 1.00
#> row_norm: <lgc> TRUE
#>    maxit: <int> 100
#>      tol: <nmr> 1e-04