Creating a Random Distribution on Every Run

[johnjohnson4]

Hello,
Is there a way to have a random distribution change on each simulation run?

I am doing a sensitivity analysis and would like to have the random distribution (preferably uniform or normal) change on each run. Can I make the seed random?

Thanks,
john

[Administrator]

You can using GET TIME VALUE (and measure 10)
http://www.vensim.com/documentation/ind ... random.htm

Sensitivity simulations should automatically select from a distribution though. Or are you doing many sensitivity runs using different RN seeds?

[johnjohnson4]

Correct. I am trying to do many sensitivity runs and would like to use a different seed each time. This will give me a broadly distributed residual error for my data points. Looks like GET TIME VALUE will get different seeds within the simulation run but each run will still be the same. Correct?

I need to be able to have a randomly selected seed on each simulation run.

[johnjohnson4]

Tried GET TIME VALUE. Its not supported in PLE+

Any other options?

[Administrator]

That's the only thing I can think of.

[Administrator]

How many seeds do you need to run the sensitivity test with?

[LAUJJL]

Hi

Use the model joined.

It is necessary to change the formulation of the seed variable to 0 for the first run, as the get vdf function does not return the atttime if the file is empty. All the successive runs will have the seed of the distribution automatically changed. The get vdf constants function should work with the ple plus, at least the documentation says so.

Regards.

JJ

[tomfid]

Very clever, JJ.

If there's demand, we could potentially introduce a new convention for randomizing the streams - perhaps a negative seed value?

[LAUJJL]

Hi Tom

I do not see the utility of using negative values for seeds.

But it would be useful if the get vdf constants could produce the attime value as a result, if the file is empty or does not exist to avoid the necessity to run a first model to generate a file to start with.

Regards.

JJ

[tomfid]

The purpose of the negative seed would be to serve as a flag indicating that results should be randomized (to the clock, or to system crypto sources) with every new run.

I'll take a look at the vdf code.

[johnjohnson4]

JJ!

Thanks for the get vdf idea. I'll give that a try.

I actually figured out a solution as well. If you make the seed a variable and set it as a multivariate parameter in the sensitivity analysis, you will get a different seed on each simulation run.

I'll add your get vdf to my bag of tricks.

Thanks again,
john

[tomfid]

You can also use stream 0 for all the function calls, then set the global NOISE SEED parameter with the sensitivity control file.