MC toys parallelization

[Jacopo]

Hi all,

I am using the RooMCStudy class for toy studies. Is there a way to generate and fit the toys using multiple cores in parallel?

Thanks,
Jacopo

[StephanH]

Hi @Jacopo,

not directly, but you can use

RooMCStudy(model, observables, FitOptions(..., NumCPU(xx), ...) )

to at least run the fit in parallel.

Then, you can run the process multiple times from a script with different random seeds, and save the fit results (FitOptions( ..., Save(), ...)). You can gather all the fit results from different files, and aggregate them into the one study using addFitResult.

Then, there is also the poorly documented RooStudyManager.
It can apparently run a ToyMCStudy in parallel, but it’s not exactly doing the same thing as the RooMCStudy.