choice-metrics.com

Hi

I have a question regarding the blocking functionality. I would like to be able to reproduce a blocking structure so that it won't change every time I open the design file. And to know more about what an appropriate search time is.

In a previous post Johnr have written "To locate a blocking column for a non-orthogonal design, Ngene randomly searches through blocking columns and attempts to minimise the correlation between the blocking column and the design columns. Given that it will use different seeds, it will locate different blocking columns each time, unless you use a larger number of iterations in the sorting (which at some point converge to the same solution). You can change the default values in the blocking column search. This is described in Section 11.3.4 of the manual."

However, is it correctly understood that the seed cannot be defined in order to make sure that the blocking structure does not change with each re-opening of the design?

To be sure that the blocking structure is appropriate (near-minimum correlation) - and possibly obtain convergence to the same solution - I tried increasing the search time. It appers that 55 seconds was the maximum NGENE could handle (without going into indefinite non-responding mode) - this was not enough to acheive convergence to identical blockings. But it is considerably longer than the default of 3 seconds. Is longer time better, or is the default of 3 seconds ok?

Looking forward to your response, thanks!

Hi

The current behaviour of Ngene is probably not ideal. Once a blocking column has been generated, there is no need to generate it again. It should remain the same if a design is opened, closed, then opened again, in the one session. If it is saved to a .ngd file, then it will be generated again when it is opened. The original blocking is stored in the saved design however. So perhaps a workaround it to manually retrieve the blocking from the file. Right click on the saved design in Windows, and open with a text editor. You will see the design and the syntax. The final column of the design is the blocking column and is clearly marked.

Is this going to be okay for now, or do you need the same blocking column to be there when the design is opened again in Ngene? For the next point release, we will have Ngene leave the blocking column as is, if it has already been calculated.

Andrew

Hi Andrew

Thanks for the reply. It sounds good with the improvement for the next Ngene release - for now it is ok with the current solution.
What about the iteration timing issue? (see last part of my original question):

"To be sure that the blocking structure is appropriate (near-minimum correlation) - and possibly obtain convergence to the same solution - I tried increasing the search time. It appers that 55 seconds was the maximum NGENE could handle (without going into indefinite non-responding mode) - this was not enough to acheive convergence to identical blockings. But it is considerably longer than the default of 3 seconds. Is longer time better, or is the default of 3 seconds ok?"

Thank you!