I am doing some modeling on the heat transmission through windowpanes. The analytical method of calculating this value is described in detail in a standard. The method makes use of iterations, interpolations etc. This means that defining a single formula for the calculation of the heat transmission will be very 'messy'. So this is why I am asking: is it possible to use a script as a prediction formula? Especially I want to use the prediction profiler and the simulator.
A column formula is just a piece of JSL, So you should be able to get what you want so long as you define a column for each 'x' (input). Take a look at the attached table for an example when the function has two inputs. Here the formula is 'complex' (from fitting a Gaussian Process, actually). Inspect the formula and run the saved script.
Too add a bit to my colleague Ian's reply, going potentially one step further, once you've got the script written with the formula, if you want to 'rewrite' that script into say, SQL, or some other format, take a look at JMP Pro version 13 Formula Depot. It sounds like the native formula you are writing is perhaps one that represents a deterministic system? Might also be interesting to try some empirical models (if the data exists?) to see how well they might approximate the deterministic outcome? Could give the JMP Pro Generalized Regression platform a workout?
I'm thinking the above answers are missing the point. What I think you are saying is that you can calculate the heat transmission but the calculation is based on an iterative algorithm rather than a mathematical formula. Therefore how do you utilise visualisation tools such the profiler which rely on a formula.
I worked on a similar type of problem where I was trying to calculate equilibrium moisture conditions in a container. I had a number of inputs (physical dimensions of the container, material permeability etc) and after a number of iterations I would converge on a value for the moisture.
If you run your model with different input conditions then you can construct a table of input parameters and output responses. You now have a set of data from which you can construct a model (effectively a meta-model). You might want to try either linear regression (fit model) or nonlinear methods (nonlinear) depending on what the data looks like and how well it can be modelled. These models can then be used to generate your prediction formula, from which you can use the profiler and simulator.
If this is the case however, and you are particularly interested in the simulator, then I think you would be better off just simulating directly from your model i.e. for your inputs you use data sampled from random distributions.
I think you are right, Dave. And if the simulation is deterministic, one might also consider building a meta-model with the Gaussian Process. Coincidentally, this is how the formula in the table above came about.