bsvd: Gibbs sampler for Bayesian SVD.
In dcgerard/vicar: Various Ideas for Confounder Adjustment in Regression

Description Usage Arguments Details Value Author(s) References

This is a modification of the Bayesian approach from Hoff (2012) to allow for heteroscedastic columns. We start the missing values from the RUV4 solution.

bsvd(
  Y21,
  Y31,
  Y32,
  k,
  nsamp = 10000,
  burnin = round(nsamp/4),
  keep = 20,
  print_update = TRUE,
  plot_update = FALSE
)

`Y21`	Top left of matrix.
`Y31`	Bottom left of matrix.
`Y32`	Top right of matrix.
`k`	The rank of the mean matrix.
`nsamp`	A positive integer. The number of samples to draw.
`burnin`	A positive integer. The number of early samples to discard.
`keep`	A positive integer. We will same the updates of `Y22` every `keep` iteration of the Gibbs sampler.
`print_update`	A logical. Should we print a text progress bar to keep track of the Gibbs sampler (`TRUE`) or not (`FALSE`)?
`plot_update`	A logical. Should we make some plots to keep track of the Gibbs sampler (`TRUE`) or not (`FALSE`)?

The rejection sampler in rbmf.matrix.gibbs almost always stalls, so I am removing it from exports for now.

A list with the following elements:

Y22_array A three-dimensional array containing draws of Y22. Y22[, , i] contains the ith draw of Y22.

mu_psi_phi A matrix with three columns. The rows are the draws from the posterior. The first column is for the mean of the singular values. The second column is for the precision of the singular values. The last column is for the mean of the variances.

neff_y22 The effective sample sizes from Y22_array as calculated by effectiveSize

neff_mu_psi_phi The effective sample sizes from mu_psi_phi as calculated by effectiveSize

David Gerard

Hoff, P. D. 2012. "Model averaging and dimension selection for the singular value decomposition". Journal of the American Statistical Association. doi: 10.1198/016214506000001310

dcgerard/vicar documentation built on July 7, 2021, 1:08 p.m.