steepest descent/ascent

[#1]

Dear JMP community users,

I would like to How can I perform steepest ascent/descent using JMP. Does JMP has an Option for that.

[#2]

I think that you should be more specific. Steepest ascent/descent belongs to mathematical optimization techniques, e.g. see this wiki Gradient descent .

Depending on the problem in your hands, you may not need to implement the method by yourself, because JMP may already have capabilities for your problem. If you don't know whether such capabilities exist in JMP, you can try to elaborate your problem description. If indeed, there is no existing capabilities out-of-the-box, JMP have powerful tools to solve your customized problem if you are able to specify the problem and tell JMP. One of such tools is the Nonlinear platform, using which you don't need to implement your own method. In the end, JMP has a programming language, JSL, using which, you might be able to write a program which does the mathematical optimization. The thing that you need is the ability to translate a mathematical optimization method to a program.

[#3]

Is ascent/descent meant from timestep to timestep?
You can calculate the slope via :delta= :col - lag(:col).Then you just have to search for col maximum (:delta) or minimum (:delta).

Maybe the raw data is affected by some noise - then col maximum might be just a noisy spikes.

In this case, it will help to use a smoother and save the formula. Then you could find the large slopes analytically ...

[#4]

My interest towards these methods is to find out valuable results from DOE which we have performed like let's say we have performed two DOEs (All significant factors shows Linear effects), now I would like to know where my significant factors has quadratic effect from my trails such that I can go in that direction with further trails or DOE.

This is what my understanding about these methods.

[#5]

I am not a specialist in DOE, but you have described your objective clear enough to me, and I hope other experts in the area can help.

From my personal attempt, I think this page in the documentation might be relevant, though not specifically suggesting only looking at quadratic terms.

Example of a Response Surface Design

[#6]

Steepest ascent is a method to move the design space.  I think you are interested in augmenting the design space you are in.  There are a number of ways to do this.  Peng points to one area you look at.  Here is another.

https://www.jmp.com/support/help/en/17.2/?os=mac&source=application#page/jmp/augment-designs.shtml

[#7]

For a numerical example, you can refer to the following article:

George E. P. Box & Patrick Y. T. Liu (1999) Statistics as a Catalyst to Learning by Scientific Method Part I—An Example, Journal of Quality Technology

You are also going to find another good example in Montgomery's book Design and Analysis of Experiments.

Please let me know if you need any further information,

[#8]

My understanding of steepest ascent comes from Box.  His thinking is the use the information gathered from an initial experiment to develop a model (theoretically perpendicular to the contours of the response variable).  Use this regression model to run "tests".  Use about the same level setting spacing that was used in the initial experiment.  Run those tests in a sequence that changes the levels of the factors that are in the model.  Keep running them as the response increases, until there is a point of inflection (the response goes down).  Move the next experiment to before the point of inflection.

JMP will be able to give  you the model.  You will have to do the initial experiments and the subsequent testing.

This can be useful as long as you are not too far from optimum (e.g., factors should be continuous vs. categorical).