SuperPCA (version 0.4.0)

SIFA: Supervised Integrated Factor Analysis

Description

Supervised Integrated Factor Analysis

Usage

SIFA(
  X,
  Y,
  r0,
  r,
  relation = "linear",
  sparsity = 0,
  max_niter = 1000,
  convg_thres = 0.01,
  type = "Anp"
)

Arguments

X

n*q matrix, centered covariate data

Y

1*K cell array, each cell is a n*pi centered primary data, should roughly contribute equally to joint struct

r0

scalar, prespecified rank of common structure

r

1*K vector, prespecified rank of specific structures

relation

'linear' (default), use linear function to model the relation between U and covariates 'univ_kernel', use kernel methods for single covariates

sparsity

1 (default), when est B0 or B, use LASSO with BIC to select the best tuning, suitable for high

max_niter

default 1000, max number of iteration

convg_thres

default 0.01, overall convergence threshold

type

"A" or "Anp" or "B" or "Bnp", condition to use

Value

list with components

B0:

q*r0 matrix, coefficient for joint structure (may be sparse)

B:

1*K cell array, each is a q*ri coefficient matrix (may be sparse)

V_joint:

sum(p)*r0 matrix, stacked joint loadings, with orthonormal columns.

V_ind:

1*K array, each is a pi*ri loading matrix, with orthonormal columns

se2:

1*K vector, noise variance for each phenotypic data set

Sf0:

r0*r0 matrix, diagonal covariance matrix

Sf:

1*K array, each is a ri*ri diagonal covariance matrix

EU:

n*(r0+sum(ri)) matrix, conditional expectation of joint and individual scores

Examples

# NOT RUN {
r0 <- 2
r <- c(3,3)
V <- matrix(stats::rnorm(10*2),10,2)
Fmatrix <- matrix(MASS::mvrnorm(n=1*500,rep(0,2),matrix(c(9,0,0,4),2,2)),500,2)
E <- matrix(stats::rnorm(500*10,0,3),500,10)
X <- tcrossprod(Fmatrix,V)+E
X <-scale(X,center=TRUE,scale=FALSE)
Y1 <- matrix(stats::rnorm(500*200,0,1),500,200)
Y2 <- matrix(stats::rnorm(500*200,0,1),500,200)
Y <- list(Y1,Y2)
SIFA(X,Y,r0,r,max_niter=200)
# }