topic Nonlinear fit using loss function with penalty in Discussions

Nonlinear fit using loss function with penalty

LucaC — Fri, 16 May 2025 09:23:35 GMT

HI All,

I need to do a nonlinear fit imposing a constraint equation to the parameters besides the range bounds. To do that, I am using the nonlinear platform defining the loss function with penalty. I wrote the model formula in a column where I defined the fitting parameters and in an other column I defined the loss function I am trying to use. Below I report a description of the loss function:

(:Y-:model)^2+lambda*(a - a1*exp(-b1/b2)-....-exp(-b1/b5))^2

a,a1,b1,b2,...,b5 are the fitting parameters defined in the model formula column. When I lunch the platform I obtain this message:

Does this mean it is not possible to use a loss function with penalty or am I doing something wrong? Otherwise, how could I implement fitting with constraint equation for the parameters in JMP? Thanks in advance for your help.

Re: Nonlinear fit using loss function with penalty

peng_liu — Fri, 16 May 2025 17:23:39 GMT

I don't think that is related whether the formula has a penalty or not. But you mentioned that a, a1, ... are in Loss column, but defined in Model column, that does not look workable to me. Can you write the entire thing as a single formula, i.e. without the middle step of creating "Model" column?

Such as, instead of write Model using a formula like xyz, you write a Loss column like this:

Parameter({a=1, a1=2, ...}, (Y - xyz)^2 + lambda * (..... )^2 )

And then you feed this column directly into Loss.

Re: Nonlinear fit using loss function with penalty

MathStatChem — Fri, 16 May 2025 18:04:54 GMT

see this discussion: https://community.jmp.com/t5/Discussions/Nonlinear-platform-custom-loss-function-shrinkage-estimation/m-p/843984