Absurd results when using multiple cores about multibugs

[rjbgoudie]

Thanks for the report.

Are you able to provide code (model file, data, inits) that produces the problem? Ideally, a minimal, simple example. This will make it easier to debug your problem.

[student982]

Thanks for your quick response.
Here are the model specification and initials values for both chains (first file) and data for 100 (out of 10000) observations (second file).
ModelTest100.txt
DataTest100.txt

[rjbgoudie]

Closing, fixed by 7dee319. Thanks again @student982 for nice and simple example of the bug :)

The problem was in GraphBeta. MultiBUGS was not externalizing / internalizing the field called normalize. This field asks the beta density to calculate normalizing constants and is needed when the beta is used as a likelihood as in this model.

Results now match up to MC error between OpenBUGS and MultiBUGS.

MultiBUGS (2 chains, 4 cores, 10000 burnin, 50000 post burn):
mean median sd MC_error val2.5pc val97.5pc start sample ESS
b0 0.6824 0.6831 0.1659 0.002433 0.3596 1.01 10001 100000 4648
b1 -0.1971 -0.1954 0.1067 8.251E-4 -0.4099 0.00761 10001 100000 16725
b2 -0.044 -0.04424 0.03268 4.768E-4 -0.1078 0.02051 10001 100000 4699
b3 0.1236 0.1209 0.2358 0.002199 -0.3344 0.5949 10001 100000 11493
b4 -0.3708 -0.3724 0.2815 0.002352 -0.9246 0.1917 10001 100000 14321
phi 4.202 4.176 0.5423 0.00216 3.22 5.333 10001 100000 63018

OpenBUGS (2 chains, 10000 burnin, 50000 post burn):
mean sd MC_error val2.5pc median val97.5pc start sample
b0 0.6845 0.1663 6.479E-4 0.3553 0.6849 1.01 10001 100000
b1 -0.1978 0.1068 1.979E-4 -0.4107 -0.1969 0.008764 10001 100000
b2 -0.0445 0.03244 1.347E-4 -0.1082 -0.04478 0.01946 10001 100000
b3 0.1215 0.2338 0.001327 -0.3292 0.1185 0.588 10001 100000
b4 -0.3682 0.2795 0.001917 -0.9122 -0.37 0.1844 10001 100000
phi 4.204 0.5427 0.001802 3.212 4.177 5.334 10001 100000