Turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- I want a space filling design with some discrete or categorical factors. Would i...

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

Highlighted
##
I want a space filling design with some discrete or categorical factors. Would it be wise to quantize or make logic decisions in my test matrix based on the value of continuous factors?

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Apr 22, 2015 1:44 PM
(2676 views)

I want a space filling design with some continuous and discrete of categorical factors. Is it wise to quantize continuous factors or create logic in my experiment based on the values of continuous factors to achieve this effect?

3 REPLIES 3

Highlighted
##
Re: I want a space filling design with some discrete or categorical factors. Would it be wise to quantize or make logic decisions in my test matrix based on the value of continuous factors?

Binning a continuous is certainly one way to do this. If you don't have too many categorical factors/levels, JMP 12 now allows categorical factors in Space Filling for Fast Flexible Filling Designs (Fast Flexible Filling Designs), so you may want to try that out.

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Highlighted
##
Re: I want a space filling design with some discrete or categorical factors. Would it be wise to quantize or make logic decisions in my test matrix based on the value of continuous factors?

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

These FFF fast flexible filling designs have a problem that I imagine stems from the clustering technique used to generate them. My experience so far is that they don't go all the way to the edge of the factor space. The picture I have in my mind is that the clusters would be bounded by the factor space and the final design point from the cluster would be in the interior of the cluster. This would lead to a buffering along the boundaries of the factor space and leave out the extreme points. Am I doing something wrong?

I've had another problem too. I added a "Disallowed Combination" that completely drives the algorithm crazy. It generates all manner of combinations outside the bounds defined when I added my continuous factors. I haven't even gotten to the point where I am binning the continuous values. I'm running version 11.1.

Highlighted
##
Re: I want a space filling design with some discrete or categorical factors. Would it be wise to quantize or make logic decisions in my test matrix based on the value of continuous factors?

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

One of the new features in JMP 12 was to add the criterion of MaxPro (which is also the default) that does a better job at exploring the boundaries.

There was a bug in JMP 11.1 for non-convex regions with disallowed combinations if the centroid fell outside of the design range, that was fixed in JMP 11.2, so it's quite possible that you've run into that case.