Closed sigmafelix closed 3 months ago
I apologize; I am just seeing this. (In the future, feel free to tag me and/or e-mail me directly if you run into issues like this.) Did you see what examples is causing this? My suspicion is that this is an issue with GPvecchia, because I have had similar issues trying to run it on our Linux servers. As a workaround, try running the command "export OMP_NUM_THREADS=1" (in bash shells) before starting R. Hopefully that will fix it.
If you really want to try to debug this, you could try running the example both as a univariate model and as a multivariate model of length 1. If my suspicion is correct, the latter will work but the former will not. I believe GPvecchia uses OpenMP for parallelization, but we don't do any kind of parallelization in our multivariate version of the Vecchia approximation. If that's the case, then I don't think there is anything we can do to fix it on our end. But if the workaround describes above does the job, that's fine, too.
@ericbair-sciome Thank you for the comment. I find setting export OMP_NUM_THREADS=1
before opening R does not make any OpenMP related errors.
I installed PrestoGP in
main
with commandremotes::install_github("Spatiotemporal-Exposures-and-Toxicology/PrestoGP")
on HPC (Rocky Linux x86_64).When running examples, R session crashed with error messages
I tried
export USE_OPENMP=1
before opening R session then reinstalled PrestoGP, but it didn't work. The HPC node has Rocky Linux openblas-openmp.x86_64 and openblas-openmp64.x86_64, so I don't think this error is related to openblas installation.