flash_hd: Factor Loading Adaptive Shrinkage (heteroscedasticity...

In NKweiwang/flash: Title Case

Description

flash provide rank one matrix decomposition

Usage

flash_hd(Y, tol = 1e-05, numtau = 500, partype = "constant",
  sigmae2 = 1)

Arguments

Y	is the data matrix (N by P)
tol	which is the stop criterion for the convergence, default is 1e-5
numtau	number of iteration, default is 500. for the backfitting case, the number of tau should be 5 or 10.
partype	type for the noise variance, which takes values from "constant": constant variance, "anova": variance is from anova model, "loganova": log variance is from anova model, "poisson": variance is equal to mean, else user should provide known value for sigmae2
sigmae2	which is a matrix for the known noise variance matrix. the default is 1 which is not useful

Details

flash_hd privide rank one matrix decomposition with variational EM algorithm.

Value

list of factor, loading and variance of noise matrix

• l is a N vector for loadings

• f is a P vector for factors

• sigmae2 is mean of sigma square which is estimation for the noise variance

Examples

N = 100
P = 200
Y = matrix(rnorm(N*P,0,1),ncol=P)
g = flash_VEM(Y)

NKweiwang/flash documentation built on May 7, 2019, 6:02 p.m.