@andre.pfeuffer thanks for the model. I wrote a parallel version of it and ran a few benchmarks so we can add that to the collection here.
bpl.data.R (6.1 KB) bpl.stan (1.8 KB) bpl_parallel.stan (2.0 KB)
The base model has 1000 data points. To make the model slower I added a ‘rep’ argument to the data file. rep=2 means double the data length by replicating it twice.
rep = 10
grainsize = 1250
8 threads parallel 88s
1 thread parallel 179s
1 thread serial 214s
rep = 10
grainsize = 125
8 threads parallel 60s
1 thread parallel 171s
1 thread serial 214s
rep = 1
grainsize = 125
8 threads parallel 7.7s
1 thread parallel 18s
1 thread serial 18s