fisabio / pbugs

Parallelized version of the R2WinBUGS library of R

pbugs

pbugs is an R package devised to run WinBUGS or OpenBUGS models in parallel. Each chain is run in a different BUGS instance so it is computed in a different core of your machine. Ideally, if n.chains are simulated (and your simulation is long enough) this should yield computing time improvements close to 1 / n.chains.

The pbugs package heavily depends on the R2WinBUGS and R2OpenBUGS libraries. Indeed, any R syntax calling to any R2WinBUGS or R2OpenBUGS should work equally when using the pbugs package by simply replacing any bugs(...) sentence by the corresponding pbugs(...) sentence. We have tried to preserve the original syntax so that your old code using this package works also when the pbugs package is used instead, by simply changing bugs by pbugs. Anyway, when pbugs is used that same code should run substantially faster.

Aditionally, the pbugs package contains also some improvements over R2WinBUGS and R2OpenBUGS that we have thought interesting to implement as former R2WinBUGS users. These main improvements could be summarized as:

• pbugs objects keep the structure and functionality of bugs objects with some enhancements, so anything that you did with your bugs will be also possible to do with your pbugs object.
• pbugs automatically saves the computing time taken to run the sentence generating that object. This is saved at the exec.time component of any pbugs object.
• pbugs makes it possible to change and control the default sampling methods when using WinBUGS (as described at the ‘Changing MCMC defaults’ section of the WinBUGS help) automatically.
• If some of the WinBUGS chains crashes during the MCMC simulation process for some reason, pbugs will retrieve the results of the rest of chains and will build the corresponding pbugs object. In this manner you will not lose the results for the rest of chains.
• Improved functionality for print and summary methods for pbugs objects, allowing for example ordering variables in summaries by their values of their convergence statistics.
• An enhanced history plot function for pbugs objects (traceplot) has been included into the package. This function automatically plots those parameters with deficient convergence criteria.

pbugs has been substantially tested on Linux workstations, we have also substantially tested the package on Windows machines, but although tested, our experience on ‘Mac’ systems is much lower. Thus, if you found some issue on the running of this package on some of these different environments please do not doubt to let us know*.