You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Describe the bug
The nruns argument of the mcmc() function is supposed to specify the number of independent replicates. This works fine with the single-thread version of RevBayes, but if the analysis is executed using the MPI version, some of the replicates are apparently initiated with the same random seed. This makes them draw exactly the same samples, producing redundant output and reducing the number of independent replicates below that specified by the user.
To reproduce This zipped directory contains a script specifying 10 independent runs and all the data required to run it. One absolute path has to be changed in the script. When I execute it using the MPI version of RevBayes with one core per run (mpirun -np 10 ./rb-mpi), the analysis will indeed start with 10 runs, but of these, no more than 3 to 7 will actually be independent of one another.
Expected behavior
The replicates should all be independent of one another.
Computer info
macOS 14.3 (Sonoma), RevBayes 1.2.2, latest version of the development branch (rapture-2386-g54c6ba) and Red Hat Enterprise Linux 8.4 (Ootpa), RevBayes 1.2.2, development branch, commit rapture-2364-ga6a40c.
The text was updated successfully, but these errors were encountered:
Describe the bug
The
nruns
argument of themcmc()
function is supposed to specify the number of independent replicates. This works fine with the single-thread version of RevBayes, but if the analysis is executed using the MPI version, some of the replicates are apparently initiated with the same random seed. This makes them draw exactly the same samples, producing redundant output and reducing the number of independent replicates below that specified by the user.To reproduce
This zipped directory contains a script specifying 10 independent runs and all the data required to run it. One absolute path has to be changed in the script. When I execute it using the MPI version of RevBayes with one core per run (
mpirun -np 10 ./rb-mpi
), the analysis will indeed start with 10 runs, but of these, no more than 3 to 7 will actually be independent of one another.Expected behavior
The replicates should all be independent of one another.
Computer info
macOS 14.3 (Sonoma), RevBayes 1.2.2, latest version of the
development
branch (rapture-2386-g54c6ba) and Red Hat Enterprise Linux 8.4 (Ootpa), RevBayes 1.2.2,development
branch, commit rapture-2364-ga6a40c.The text was updated successfully, but these errors were encountered: