Prediction formula options

Peter_Hersh

This customer wants a very specific output for his prediction formula from a model shown below.  He has 11x factors and a constant c1 and his response y.  He is trying to get numbers for f1,f2.....,f10 to fit his y.  Does anyone have any thoughts on if or how this could be done?

y = f1(x1,x2,x3,x4) * sign(x2) + f2(x1,x2) + f3(x1,x3,x4) * c1 * x5 +

     f4(x1,x3,x4) * c1 * x6 + f5(x1,x7,x3,x4) * sign (x7) + f6(x1,x8,x3) * c1 * x7 +

     f7(x1,x2,x8,x3) * sign(x2) + f8(x1,x8,x2,x3,x4) * x9 + f9(x1,x3,x10,x4) * x10 +

     f10(x1,x2,x3,x4,x8,x11) * x11

Thanks,

erich_gundlach

Pete,

The nonlinear platform can do that. (I used to do the same thing to fit data to pre-defined reaction rate models.) How well it works depends a bit on the starting values you give JMP for the coefficients. I never really messed with the other settings you can tweak, as the defaults always worked fine. It always did a very good job for me, assuming the pre-defined model was appropriate for my data.

Erich