Parallelize `bf()`

[JeffreyRStevens]

Does the bf() function have a way to parallelize the computations across repetitions? Or should I run bf() repeatedly using external parallelization? Basically, I'm looking for any way to speed up processing for a very long processing time.

[quentingronau]

Hi Jeffrey,

The bf function does not, but the bridge_sampler function has. Once the log marginal likelihoods have been computed via bridge_sampler, using bf should be very quick.

Cheers, Quentin

[JeffreyRStevens]

Many thanks for your reply, @quentingronau. That clarifies where the action is. When using the bridge_sampler() function, I see that I can specify the number of cores to parallelize the processing. I have a few questions, if you don't mind.

• Does the parallelization work between or within repetitions?
• How can I best speed up processing? Currently, if I run on 1 core, one of my models takes 14.6 minutes. Bumping up to 4 cores drops the time to 10.3 minutes and 8 cores is 9.7 minutes. So I'm not getting a huge benefit from many cores. Also, in watching my system resources, multiple cores are only used a relatively small part of that processing time (less than half). Are there any other ways to speed up processing besides increasing cores?

System specs: R version 4.2.2; bridgesampling version 1.1-2; Ubuntu 22.04.2