Subscribe Bookmark RSS Feed

Custom Design - I-optimal - center points not added

joberi

Community Trekker

Joined:

Jun 23, 2011

Hi All,

I just tried to design an I-optimal Experiment with 7 continuous factors with 48 runs. Within I'd like to have 6 center points. However, JMP does simply not include those, despite me entering those explicitly in the "Number of Center Point" field.

Any ideas?

Thanks!

11 REPLIES
joberi

Community Trekker

Joined:

Jun 23, 2011

just a quick add on:
it seems to be related to the sequence of ticking boxes (also concerning blocks) and adjusting the number of runs. I just tried once more and realized that under particular sequence, it seems to work. Which one that is, I could not find out and would appreciate guidance.
Thanks!
louv

Staff

Joined:

Jun 23, 2011

Your attachment indicates that you specified "6 replicate runs" rather than "6 center points"

I added 7 continuous factors with an RSM Model and then chose I-optimal criteria and  6 center points and entered 48 runs.

Screen Shot 2016-11-18 at 8.26.49 AM.png

 

 

joberi

Community Trekker

Joined:

Jun 23, 2011

Hi Louv,

thanks for your answer - that was just a mistake when I copied it in. It actually only appears that the center points don't come up if I have Blocks.

Best regards

M_Anderson

Staff

Joined:

Nov 21, 2014

Hi!  I just tried to replicate what you described (including the blocking) and the center points came up.  Could you generate the table and post it so that we can see the set up you're using and the resulting table?

 

Thanks!

 

M

joberi

Community Trekker

Joined:

Jun 23, 2011

custom_design_.jpg

So that is it - first, I go "Custom Design", I put in 7 contiuous factors, click ok, then click "RSM", add a tick to the blocks and type 24 into it, add "6" to the Center Points and then specify 48 for the Number of Runs.Then make design and get the table as shown. Only 3 center runs are there now.

Thanks!

louv

Staff

Joined:

Jun 23, 2011

Not sure why you are not getting your desired number of center points. I was able to get all 6 center points although I often have to start the platform from scratch and re-enter my settings to obtain the design I am looking for with respect to balancing the center points within the block. Also there are times when I would simply just have a design for N=42 in your case and manually add my 3 center points per block for a total of 48 runs to get the desired scenario since the optimality criterion will at times result in an unbalance the number of center points per block.

Just out of curiosity have you investigated the use of the Definitive Screening platform for this scenario. The number of runs would be considerably less. If you already know that the factors are active and screening has already been completed than you can ignore this last question.

 

Lou

 

Screen Shot 2016-11-18 at 11.52.08 AM.png

 

 

Screen Shot 2016-11-18 at 11.52.48 AM.png

 

 

 

joberi

Community Trekker

Joined:

Jun 23, 2011

Hi Lou,

thanks for your answer. Well actually adding manually is my workaround, but I was wondering, if the design would be somehow "better", if JMP already includes my center points up-front. Or is that anyways not the case?

Concerning the Design itself - yes, I also use Defnitive Screening Designs (and I like them), but in this particular case - I know that the factors most likely will be significant and I want to avoid iterations.

Best regards

louv

Staff

Joined:

Jun 23, 2011

That's what I sort of suspected.

In any event, even when I increase my number of starts to 1000 and indicate 6 center points with a block size of 24 and 48 runs the custom design afforded me 2 center points for Block 1 and 4 center points in Block 2. (Rows 19,22,26,29,34,43)

Screen Shot 2016-11-18 at 1.17.31 PM.png

 

 

David_Burnham

Super User

Joined:

Jul 13, 2011

I followed your steps using JMP 13 on WIndows and got the centre points as expected.

 

Regarding whether it is better to include centre points when you generate the RSM rather than after: It is "better" in the sense that the algorithm will take account of the presence of the centre points (that it true for recent versions of JMP, it wasn't true for earlier versions). But since I-optimal designs don't really derive much benefit from centre points adding the points after the design is generated will have neglible impact.  I created designs using both routes and did a design comparison ('custom design' had centre points added at the time of generating the design, and with 'custom design 2' I added the centre points myself after the design had been created). 

 

design-efficiency.PNG

 

-Dave