STRS: Self-tuning Rank Selection.

In bingx1990/STRS: Self-tuning rank selection in multivariate response regression

Description

STRS estimates the reduced-rank of the coefficient in multivariate linear regressions. It can also be used to select of the number of factors in factor models.

Usage

STRS(
  data_Y,
  data_X,
  type = "STRS-MC",
  rank_X = NULL,
  self_tune = TRUE,
  rep_MC = 200,
  rank_tol = 1e-04,
  C = 2.01
)

Arguments

data_Y	A n by m response data matrix.
data_X	A n by p feature data matrix.
type	String. One of \"STRS-DB", "STRS-MC" and "SSTRS"\. The default is "STRS-MC". "STRS-DB" and "STRS-MC" are for general dimensional settings; "STRS-DB" uses deterministic expressions for updating lambda while "STRS-MC" updates lambda by Monte-Carlo simulations. "STRS-DB" is recommended if "STRS-MC" is computationally burdensome. "SSTRS" is a simpler version of "STRS-DB" when either n >> m or n << m.
rank_X	An integer, the specified rank of data_X. If unspecified, this is estimated from data_X as the largest k such that σ_k(data_X) ≥ rank_tol.
self_tune	Logical. TRUE if iteratively estimate the rank and FALSE otherwise. The default is TRUE.
rep_MC	An integer. The number of Monte Carlo simulations. Default is 200.
rank_tol	The tolerence level for determining the rank of data_X when rank_X is NULL. Default is 1e-4.
C	A numerical constant for the intial lambda. Default is 2.01.

Value

The estimated rank.

Examples

  library(STRS)
  est_rank <- STRS(Y, X)
  est_rank <- STRS(Y, X, type = "STRS-DB")

bingx1990/STRS documentation built on Jan. 22, 2022, 8:46 a.m.