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
- :
- How do I evaluate a JMP derived desirability function on specific parameters not...

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

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

Jun 14, 2018 9:02 AM
(1547 views)

Context: I have data from a respose surface experiment. I've modeled the responses and obtained the desirability function (e.g. I used the profiler to "save desirabilities" got a new column in my table and can double click the column header and see the function). I know about as much as one can about how that function comes about up to what I think is proprietary --- that it is Exp(w_1Log(d_1) + w_2Log(d_2) + ... + w_k Log(d_k)) where each d_i is a specific desirability function for a specific response, e.g. a smooth piecewise function that is 2 exponentials and a cubic for minimize/maximize and a couple of scaled normal densities for targets.

Question: I want to evaluate the Desirability on a set of parameters not run in the experiment --- just like JMP already does for finding optimal parameters, e.g. it uses algorithms to move across the the parameter space, uses the models to predict response values for parameter combinations and then computes the desirability of that combination. I want to specify parameters and have it compute the predicted responses and desirability for a particular set of models I've fit. Can I do this? If yes how? One option would be to do it myself since I have the model parameters I can write those models down, get predicted values and then plug those into the desirability function, but then I would need to have the specific smooth piecewise functions and that seems hard given that the details are proprietary (or seem to be and if I was JMP I would make them that way). Any ideas?

1 ACCEPTED SOLUTION

Accepted Solutions

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

The Prediction Profiler is conveniently available in the fitting platforms. It is also available on its own from the Graph menu. As such, it uses a saved model (column formula). So you just have to create a new column with a formula with the know relationship to the same response variable. Select Graph > Profiler, select the data columns with the fitted model and the known model, assign them both the Y role, and click OK.

Learn it once, use it forever!

1 REPLY 1

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

The Prediction Profiler is conveniently available in the fitting platforms. It is also available on its own from the Graph menu. As such, it uses a saved model (column formula). So you just have to create a new column with a formula with the know relationship to the same response variable. Select Graph > Profiler, select the data columns with the fitted model and the known model, assign them both the Y role, and click OK.

Learn it once, use it forever!