191893_en-US

liliconcom

Web Content Suggestion Request - Analyzing Simulated Data

Hi!  I got a request from a customer for a webinar recommendation.  He's in a situation where he has data from a first principles simulation he's wanting to build a profiler, bring in some physical uncertainties in the factors, and then use the Simulate and Variable Importance tools to explore the data.

 

That seems like a fairly straight forward workflow - but I can't find any whitepapers or video content demonstrating it.  

 

Does anybody have anything that might fit the bill?

 

Thanks!

 

M

Re: Web Content Suggestion Request - Analyzing Simulated Data

Hi @MikeD_Anderson,

 

If I understand the question correctly, they would only need a formula for the response and add it as a column. All of the X's would also have to be added as columns as well.  Then they only need two rows: 1) Min of all X's 2) Max of all X's.

 

Example is like this one attached (Time to get home).

 

Is this what you were looking for?

 

Chris

Re: Web Content Suggestion Request - Analyzing Simulated Data

@chris_kirchberg - right idea - do you know of any webinars for setting this kind of thing up?

Re: Web Content Suggestion Request - Analyzing Simulated Data

None that I can think off the top of my head.  We used to use this example when doing live events for Graphing Data or the Beyond Excel workshop.

Re: Web Content Suggestion Request - Analyzing Simulated Data

Hi Mike,

 

If I understand you correctly, I did work like this in a former life.  Happy to discuss when I get back home next week.  Basically the simulation data is deterministic, but there is bias.  You can see that in Actual vs. Jackknifed Predicted in Gaussian Process.  It's more complicated than I making it sound, but that is a good place to start.  One thing I found though is that NN models are usually better for these types of simulations because they are typically highly non-linear. 

 

Above is the modeling step.  You would want to start things off by using a space filling design based on the first principles equations and inputs used for the "simulation experiments".  This results from these "experiments" are then used to build your approximation/surrogate model.     The rule of thumb for the number of runs in the space filling design is 10X the number inputs.  JMP Pro has a FFF - fast filling flexible design that is typically better than the standard space filling designs and it also handles constraints fairly well.

 

Let me know.

 

Best,

Bill