Same random seed in gridpack

Asked by Baptiste Ravina on 2020-04-09

Hi, having generated a gridpack and running
./run.sh <nevents> <iseed>
I observe that two consecutive executions give rise to different outputs. I realise this might indeed be the intended behaviour, in order to generate independent datasets each time. However in my particular case I need to keep track of the random numbers used each time.
How can I modify the contents of the gridpack such that running
./run.sh <nevents> <iseed>
twice generates the same output, but
./run.sh <nevents> <iseed>
./run.sh <nevents> <iseed+1>
doesn't?
Effectively, how do I make sure the only random seed ever used in the generation of events is the iseed?

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Solved by:
Olivier Mattelaer
Solved:
2020-04-13
Last query:
2020-04-13
Last reply:
2020-04-12

Which version of the code are you using?

Cheers,

Olivier

> On 10 Apr 2020, at 00:47, Baptiste Ravina <email address hidden> wrote:
>
> New question #689812 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/689812
>
> Hi, having generated a gridpack and running
> ./run.sh <nevents> <iseed>
> I observe that two consecutive executions give rise to different outputs. I realise this might indeed be the intended behaviour, in order to generate independent datasets each time. However in my particular case I need to keep track of the random numbers used each time.
> How can I modify the contents of the gridpack such that running
> ./run.sh <nevents> <iseed>
> twice generates the same output, but
> ./run.sh <nevents> <iseed>
> ./run.sh <nevents> <iseed+1>
> doesn't?
> Effectively, how do I make sure the only random seed ever used in the generation of events is the iseed?
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Baptiste Ravina (bravina) said : #2

Hi Olivier, I'm using the latest version 2.7.2.

Not expected then.
I always consider than the more randomness you get the better it is but I will take a look.

Cheers,

Olivier

> On 10 Apr 2020, at 11:03, Baptiste Ravina <email address hidden> wrote:
>
> Question #689812 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/689812
>
> Status: Answered => Open
>
> Baptiste Ravina is still having a problem:
> Hi Olivier, I'm using the latest version 2.7.2.
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Hi,

Here you will find all the information that you might need:
 - a link to a patch
 - a link to a dedicated tarball including the patch
 - more
https://bazaar.launchpad.net/~mg5core1/mg5amcnlo/2.7.3/revision/296

The issue was double:
 - the argument python_seed of the run_card was not handle and therefore python code was using the default seed (which is related to the time where your code starts)
 - the restore_data script was not fully restoring the state as it should which was leading to small variation of cross-section (but not of distribution)

I guess that CMS and ATLAS never noticed the second issue since they typically start by unpacking the gridpack.
The first issue is more recent.

Cheers,

Olivier

Baptiste Ravina (bravina) said : #5

Hi Olivier,

The patch worked - thanks a lot! Will these changes be included in 2.7.3?

Cheers,
Baptiste

Baptiste Ravina (bravina) said : #6

Thanks Olivier Mattelaer, that solved my question.

Yes it will be