Improving Optimization

[JMP-Andy]

What inspired this wish list request? JMP Optimizer gives just one optimal solution with Maximize Desirability even when multiple optimal solutions may exist.  Identifying a range of solutions around the optimum would help identify an operating window and/or explore trade-offs that aren't captured in the model. 

The Desirability surface can't be viewed directly in the profiler or optimizer and must be saved as a column to visualize, explore, and set thresholds. 

To set the Desirability, there is a pop-up window for each & every response which is clunky and cumbersome.

What is the improvement you would like to see? Update the profiler/optimizer to identify multiple optimal solutions and/or a range of solutions above a specified desirability threshold. 

Add the Desirability function as a response selection option in the Profiler's Surface Plot.  

Combine Set Desirability it into a single pop-up for all responses, or embed the desirability settings for the responses in a collapsible analysis window.

Why is this idea important? More robust exploration & identification of optimal solutions in JMP.  A better user interface for Optimization efforts.

[SarahGilyard]

Thanks for posting JMP-Andy. Some of the requested items in your post are already in motion to be implemented (likely in JMP 19). We are exploring a new optimization control panel interface that will allow a much more efficient and streamlined workflow for setting desirability and performing optimization. For exploration of operating window and potential trade-offs, you might find the Design Space profiler useful. If you haven't yet tried it, might be worth checking it out to see if it accomplishes some of what you're seeking. It can be accessed in the red triangle menu from the profiler. Thanks and stay tuned! 

[SarahGilyard]

@JMP-Andy

Adding to this post...   You may already know this, but another possibility to explore the desirability surface and factor space would be to save the desirability formula to the data table as a column formula and then use it as a response in the profiler under the graph menu (remember to expand intermediate formulas). You can then use the surface or contour profiler to explore the desirability surface relative to the factor space. Agree that it would be a few less button clicks if the desirability formula was available as a response option from within the fit model platform but it doesn't add too much extra work to generate the formula column first. 

As a follow-up question, have you tried the Maximize for Each Grid Point option in the profiler? If so, would it help if this feature had a "one-click" type option to perform a grid search with each factor locked individually, then list all the points that were within a certain desirability threshold of the optimal setting? 

[PatrickGiuliano]

It might be worth noting here that the optimization control panel interface was implemented in JMP 19 and I find it particularly useful for Bayesian Optimization in JMP Pro (e.g. in changing factor ranges over which BayesOpt will search and in adding or modifying factor bound constraints), but it's also useful using the Profiler for many of it's other features since it exposes more of what's possible to the user in an organized and easier to find layout.