Which algorithm to create a random order of runs?

[#1]

Which algorithm is used in JMP to create a random order of runs on the basis of a Seed for randomisation? Can I change the used algorithm? I would like to reproduce exactly a given design to evaluate it by JMP. 

Thanks.

Sincerely

Thommy7571 

[#2]

Hi @Thommy7571,

As already answered on your similar question in the post Coding design versus numerical design, you can use the option "Set Random Seed" when creating a DoE to control the randomization of the runs. See my answer here : https://community.jmp.com/t5/Discussions/Coding-design-versus-numerical-design/m-p/767860/highlight/...

If you already have created a design in JMP, you can right-click in the script "DOE Dialog" and click on "Edit" to see the full script (with the fixed random seed used) used by JMP to generate the design.

Hope this answer will help you,

[#3]

No, that does not help me at all. I want to create the same design independent of the platform. 

And RStudio uses the Eisenbichler-Kobayashi -Algorithm, but I can change this. So, to reproduce the same design with RStudio and with JMP, I want to know which Algorithm is used by JMP. It is clear that I need to set the same randomisation seed. But this was the other question...

[#4]

Hi @Thommy7571,

Sorry if my answer sounds a bit rude, but the question was not clear. I do understand better your concern now.

However, it doesn't seem possible to create the "same" design independent of the platform, as the method used to create the design may be different, depending if you use a Classical design platform (based on predefined matrix multiplications/calculations) or Custom design platform (based on iterative algorithmic computations with the coordinate-exchange algorithm).

You may end up with the same design but since the design generation is not starting from the same point, fixing the random seed may not help you achieving the same design. The difference in the design creation and augmentation was (also !) one of my previous answer on one of your post ... : https://community.jmp.com/t5/Discussions/Optimal-Designs/m-p/765417/highlight/true#M94514 

[#5]

If I start with the same design and the same seed, the only possible difference is the algorrithm for the creation of the randomised design using the gien seed.  Of course, if different plat forms use different algorithm, than different random numbers will be produced.  I just want to know the name of the algorithm which is used to create the random numbers in  JMP. Is it also the Eisenbichler-Kobayashi algorithm?

[#6]

Hello,

I would like to add that - after a discussion with the responsible developer of the DOE-App - the support of OriginLab could give me the corresponding R Code and I just reproduced with RStudsio exactly the design they created with their App and they published as an example. However, it was important to know that to get the same design two diffferent functions could be used which create either a two factorial design or a N factorial design. So if one does not use  the same function one does not use the same type of randomisation algorithm and thus one will get get a different design with the same seed, Could you also get this information?

[#7]

This might provide some information regarding the random number generators JMP uses Using JMP > Reference for JMP Functions in Formulas > Random Functions (I'm not sure if this holds true for platforms though)

You might be able to find additional information from DoE Technical Details or References https://www.jmp.com/support/help/en/18.0/index.shtml#page/jmp/technical-details.shtml# . You can also contact JMP Support if you wish to, they will then contact (if needed) JMP developers.

[#8]

Hi @Thommy7571 : I've been sorta following this, and I'm afraid I'm a little confused (not unusual!).

Your OP is: "Which algorithm is used in JMP to create a random order of runs on the basis of a Seed for randomisation?"

So, it appears to me that you aren't asking about the design at all; you just want to know, once you have a design, how the runs are randomized. Is that correct?

If so, from an analysis point of view, it doesn't matter what the order of the runs is or what algorithm is used (so long as they are randomized such that each possible sequence of runs is equally likely). i.e., if you create a design with some random run order, then create the same design (same set of factor settings) with another random order, the designs are the same...run order doesn't enter in to it.   

Apologies if I'm missing something.   

[#9]

Yes you have been missing, that I want to reproduce the same design independentl from the platform.

This helps very much if you have a design with many runs and values which are randomised - and can't be produced easily once again in the original order - and you have many response values which need the same order of the runs in the design to get the same evaluation result. You won't attibute to each run the adequate response ten times manually - e. g. for different platforms. - and this is proned to error