nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Reproduce exact same DOE table created by another person

pgupta1646

1. Is it possible to reproduce exact same DOE table created by another person with same rows & column with same max/min (+1/-1) combination for 8 Factors/Inputs and 1 Response/Output using Custom Design in JMP 14?

2. Every time I create a DOE table with exact same order/parameters/values of Factors, the run orders for the table come different with different setting of input values/combination even with Non-Random selection. What could be the reasons? Can I reproduce my own table created few minutes ago with exact same run orders having same input +1/-1 combination for all 8 factors?

Appreciate your feedback.

Regards,

Pramod

Accepted answers

All comments

Phil_Kay

Try setting the random seed in Custom Design:

Custom DoE Random Seed.png

Custom Design uses a random search to create designs. See JMP help documentation here.

I hope this helps.

Phil

statman

Perhaps I don't understand the issue...Let's say you are creating a fractional factorial on the 8 factors (2^8-4). DOE>Custom Design. Put 8 in the Add N Factors window, Select the Add Factor drop down and select continuous. This will add 8 (X1-X8). Select continue. On the next window, select Make Design. On the next window select the Run Order drop down and select any option except Randomize ( Keep the Same, for example). The select Make Table. This will give you a JMP Table with 8 columns of X's (X1-X8) and 1 Y column and 16 rows. If you repeat this, you will get the exact same matrix every time.

pgupta1646

Statman
You got it. That is what exactly I am doing. But, I am not able to get exact same matrix when I repeat to make DOE table again. Thanks for your help anyhow.
Regards,
Pramod

pgupta1646

Hi Phil
Thanks for your response. I tried Custom Design>Set Random Seed. But I am not able to get the exact same matrix when I repeat to make DOE table again, even with 'Keep the Same' selection under Run Order drop down options.
With regards,
Pramod

Phil_Kay

Hmmm....?

I find that surprising. This might sound stupid but did you set the same random seed?

Can you attach an example of the kind of design you are creating please?

Phil

Phil_Kay

For a simple 2-level screening problem like Statman’s example I would expect you to get the same design each time. When problems get more complicated you will not get the same each time.

pgupta1646

Thanks for you quick response Phil. How do I attach any example or document?

pgupta1646

As suggested my Statman, I tried a simple way as follows;
DOE>Custom Design>Add Factor (8 Continuous: X1.....X8)>Continue>No. of Runs (Default:16)>Make Design>Run Order (Keep the Same)>Make Table. Got a DOE table.
Repeated same steps above from the scratch. I got the DOE table with different matrix, means +/- combinations of all rows were different.

Mark_Bailey

You can save a script that will give you exactly the same design every time. Click the red triangle next to Custom Design and select Save Script to Script Window the first time that you make a design. Save this script and use it the next time you want exactly the same design.

JMP automatically saves this script when you click Make Table. The DOE Dialog table script is the one that you want, but you have to save this data table to be able to use it again.

You did not tell us much about your experiment. We know that you have 8 factors. What is their nature? What effects do you expect? The definition of the model can dramatically affect the outcome (design) as @Phil_Kay explained. Also, there are situations in which there is more than one design that is equally optimal. The resulting designs might be very different to you in terms of the set of treatments but they have equal optimality criterion. JMP can't know which design you prefer. If you click Back in such a case, you will get a different design.

Actually, you will not necessarily get the same design even in this simple case. You are generating a regular fractional factorial design but you are not using the older combinatorial method that was based on group theory operations with generating rules. Custom design is based on the coordinate exchange algorithm and an optimality criterion. The algorithm starts with a random assignment of levels to all the factors in every run and then iteratively improves the design. The only way to ensure the same result is, as @Phil_Kay explained, to set the random seed right before you click Make Design.

Mark_Bailey

Addendum:

Custom design recognizes special cases such as a fractional factorial and does not use the coordinate exchange algorithm in the same way. You can see this change for yourself. Notice that in your case the result is returned immediately when you click Make Design. Click Back, change the number of runs to 15, and click Make Design. You will see a progress bar count to 10 seconds. That process is the coordinate exchange algorithm working.