Post date: Sep 11, 2019 4:24:29 AM
There is a problem with the library OpenBLAS version > 0.3.5, (at least 0.3.6) which for some Linux systems (like Fedora 30), gives
an error like
List of hyperparameters:
theta[0] = [Link sn alpha]
theta[1] = [Log precision for group]
Optimise using DEFAULT METHOD
file: smtp-taucs.c hgid: 750f3d92b580 date: Tue Sep 10 11:23:39 2019 +0300
Function: GMRFLib_factorise_sparse_matrix_TAUCS(), Line: 876, Thread: 0
Fail to factorize Q. I will try to fix it...
file: smtp-taucs.c hgid: 750f3d92b580 date: Tue Sep 10 11:23:39 2019 +0300
Function: GMRFLib_factorise_sparse_matrix_TAUCS(), Line: 876, Thread: 0
Fail to factorize Q. I will try to fix it...
file: smtp-taucs.c hgid: 750f3d92b580 date: Tue Sep 10 11:23:39 2019 +0300
Function: GMRFLib_factorise_sparse_matrix_TAUCS(), Line: 876, Thread: 0
Fail to factorize Q. I will try to fix it...
file: smtp-taucs.c hgid: 750f3d92b580 date: Tue Sep 10 11:23:39 2019 +0300
Function: GMRFLib_factorise_sparse_matrix_TAUCS(), Line: 876, Thread: 0
Fail to factorize Q. I will try to fix it...
GMRFLib version 3.0-0-snapshot, has recived error no [12]
Reason : The Newton-Raphson optimizer did not converge
Message : Condition `lambda < 1.0 / lambda_lim' is not TRUE
Function : GMRFLib_init_GMRF_approximation_store__intern
File : approx-inference.c
Line : 2947
RCSId : file: approx-inference.c hgid: 750f3d92b580 date: Tue Sep 10 11:23:39 2019 +0300
Error in inla.inlaprogram.has.crashed() :
There are two ways around this.
1. Force to load the openblas library that is in the INLA package
instead of the one on the system, by doing
cd ~/R/x86_64-redhat-linux-gnu-library/3.6/INLA/bin/linux/64bit
cp libopenblas.so.0 first
where you have to change the path depending on where the INLA package is installed.
2. Bypass the use of OpenBLAS and use the Intel MKL instead
library(INLA)
inla.setOption(mkl=TRUE)
H