modifying singular RelatednessMatrix

[alek0991]

Hi,
I have been playing with PQLseq and I have noticed so many failed convergence in cases where the RelatednessMatrix is a singular covariance matrix (e.g. GRM). The modification in place is nearPD(RelatednessMatrix, corr=T). However, corr=T should be removed or set to False as RelatednessMatrix (e.g. GRM) is sometimes a covariance matrix not a correlation matrix. Setting corr=T in cases where RelatednessMatrix is a covariance can change the matrix dramatically and lead to convergence issue. I have changed it to the following and it works well in my case.

as.matrix(nearPD(RelatednessMatrix, doSym=T)$mat)

Please let me know if I am missing anything here.

PQLseq/R/PQLseq.R

Line 69 in 0463c81

RelatednessMatrix <- as.matrix(nearPD(RelatednessMatrix,corr=T)$mat)

Best,
Ali

[gaborcsardi]

Hi, this is a read only mirror of CRAN, please see the package authors in the DESCRIPTION file. Look for Maintainer, BugReports and URL. Thanks!