Random seed

[Jean-Christophe Sapin]

Hello Harry,

Would it be possible, when we don't specify a random seed, to use a variable computer data (internal clock for example), in order to have different results at each run ?

Thank you for your answer

[Harvey Harrison]

Jean-Christophe,

Could you describe a bit more what you are trying to accomplish, I'm very vary about adding anything
that would cause the model to be non-deterministic.  We currently provide the GlobalSubStreamSeed for
Distributions to globally reseed, but that requires it to be explicitly set.

Curious,

Harvey

[Jean-Christophe Sapin]

Hello Harry,

It's interesting to run a model, several times, with different random sequences, and to measure the average and Standard Deviation of the response. We can use this for example to make a DOE

It's possible to do it now, but at each run, we have to change the randomseed of each probability distribution, and if there are lots of them, it's a little bit fastidious.

So my request is of course not to remove the randomseed (we should always have the possibility to create a deterministic model), but to add a setting, in order to change the random seed value at each run (for example, if we set the random seed to -1, at each run, jaamsim, will change the random seed, and the result will be different).

Best regards

[Harvey Harrison]

Jean-Christophe,

We've provided the 'GlobalSubStream' keyword to support this use case.  The intention is for the modeller to seed his ProbabilityDistributions
using the RandomSeed keyword to keep it distinct from other distributions in the model.  Then, you can change just the GlobalSeed each time
you want to run the same case with new random streams.

This is a bit of a strange keyword, in that there is only one value globally, so no matter which distribution you apply the
keyword to, it will take effect globally.

Harvey