bpcaNet: Bayesian PCA updates

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

Description

Perform parameter updates for PPCA using the Variational Bayes framework from Oba (2003). Not recommended to use standalone, rather it is called from within bpcapM and its wrapper pcapM.

Usage

bpcaNet(
  myMat,
  covy,
  N,
  D,
  hidden,
  numberOfNonNAvaluesInEachCol,
  nomissIndex,
  missIndex,
  nMissing,
  nPcs = 2L,
  threshold = 1e-04,
  maxIterations = 200L
)

Arguments

myMat	matrix – data matrix with observations in rows and variables in columns. (Note that this is the transpose of X in pca_full.)
covy	matrix – the unbiased sample covariance of the data matrix.
N	numeric – the number of observations.
D	numeric – the number of variables.
hidden	numeric – indices of missing values in 1:length(myMat).
numberOfNonNAvaluesInEachCol	numeric – number of observed values in each column of the data (i.e. variables).
nomissIndex	numeric – indices of rows (observations) without any missing values.
missIndex	numeric – indices of rows (observations) with missing values.
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 5 elements:

W

matrix – the estimated loadings.

ss

numeric – the estimated model variance.

C

matrix – the estimated covariance matrix.

scores

matrix – the estimated scores.

m

numeric – the estimated mean vector.

References

Oba, S., Sato, M.A., Takemasa, I., Monden, M., Matsubara, K.I. and Ishii, S., 2003. doi.

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

See Also

bpcapM, 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)
  numberOfNonNAvaluesInEachCol <- colSums(X!=0)
  nomissIndex <- which(rowSums(X!=0)==N)
  missIndex <- which(rowSums(X!=0)!=N)
  threshold <- 1e-4
  bpcaNetOutput <- bpcaNet(myMat=X, covy=covX, N=N, D=D, hidden=hidden,
    numberOfNonNAvaluesInEachCol=numberOfNonNAvaluesInEachCol,
    nomissIndex=nomissIndex, missIndex=missIndex, nMissing=nMissing,
    nPcs=nPcs, threshold=threshold, maxIterations=maxIterations)

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