ppcaNet: Probabilistic PCA updates

In HGray384/pcaNet: Probabilistic principal components analysis - covariance estimation and network reconstruction

Description

Perform parameter updates for PPCA using the Expectation-Maximisation framework from Porta (2005) and also in the R-package pcaMethods (Stacklies, 2007). Not recommended to use standalone, rather it is called from within ppcapM and its wrapper pcapM.

Usage

ppcaNet(
  myMat,
  N,
  D,
  W,
  hidden,
  nMissing,
  nPcs = 2L,
  threshold = 1e-05,
  maxIterations = 1000L
)

Arguments

myMat	matrix – data matrix with observations in rows and variables in columns. (Note that this is the transpose of X in pca_full.)
N	numeric – the number of observations.
D	numeric – the number of variables.
W	matrix – initialised loadings matrix with observed variables in rows and latent variables in columns.
hidden	numeric – indices of missing values in 1:length(myMat).
nMissing	numeric – total number of missing values.
nPcs	numeric – number of components/latent variables to use.
threshold	numeric – threshold for convergence, applied to the precision parameter tau. Updates for which the change in tau are below this threshold value stop the algorithm.
maxIterations	numeric – the maximum number of iterations to be completed.

Value

A list of 4 elements:

W

matrix – the estimated loadings.

ss

numeric – the estimated model variance.

C

matrix – the estimated covariance matrix.

myMat

matrix – the data matrix with missing values replaced by their estimated projections.

References

Porta, J.M., Verbeek, J.J. and Kroese, B.J., 2005. link

Stacklies, W., Redestig, H., Scholz, M., Walther, D. and Selbig, J., 2007. doi.

See Also

ppcapM, pcapM

Examples

set.seed(102)
N <- 20
D <- 20
nPcs <- 2
maxIterations <- 1000
X <- matrix(rnorm(50), D, N)
X <- scale(X, center=TRUE, scale=FALSE) # mean 0
covX <- cov(X)
IX <- sample(1:D, 10)
JX <- sample(1:N, 10)
nMissing <- length(IX)+length(JX)
X[JX, IX] <- 0
hidden <- which(X==0)
threshold <- 1e-4
r <- sample(N)
W <- t(X[r[1:nPcs], ,drop = FALSE])
W <- matrix(rnorm(W), nrow(W), ncol(W), dimnames = labels(W) )
ppcaNetOutput <- ppcaNet(X, N, D, W, hidden, nMissing, nPcs, threshold, maxIterations)

HGray384/pcaNet documentation built on Nov. 14, 2020, 11:11 a.m.