dica: Distributional ICA
In Emory-CBIS/DICA: Distributional Independent Component Analysis

Description Usage Arguments Value

View source: R/dica_2stage.R

Distributional independent component ICA with two-stage estimation. At stage one, a mixture model (Gaussian or Wishart) is estimated by EM algorithm. At stage two, the posterior weights (after a transformation with the multinomial logit link function) are decomposed with Infomax ICA.

dica(
  Y,
  K = 6,
  L = 3,
  Rmat = diag(L),
  tol = 1e-05,
  itr.max = 1000,
  log = FALSE,
  verbose = TRUE
)

`Y`	A numeric matrix or a list. If a matrix, a mixture of Gaussian model is to be fitted at stage one, and rows correspond to observations (`n`) and columns correspond to the dimension (`d`). If a list, a mixture of Wishart model is to be fitted at stage one. Each item of the list is a `d*d` positive definite matrix, corresponding to a observation.
`K`	The number of components at stage one.
`L`	The number of components at stage two, should be no larger than `K`.
`Rmat`	Initial estimate of the `L*L` orthogonal rotation matrix at stage two.
`tol`	Relative convergence tolerance for the log-likelihood.
`itr.max`	Maximum number of algorithm iterations.
`log`	if `TRUE`, the posterior probabilities at stage one are calculated on the log scale.
`verbose`	If `TRUE`, print progress of algorithm to console.