Choosing modelfun in fitnlm

Must a model function be supplied for all nonlinear regression problems? I have data with 8 predictors and 1 response that has no relationship (atleast none has been established at the moment) between the predictors and the response. How do I model this using fitnlm or nlinfit?

[Alex Sha]

Post out your data if possible.

Answer 1

[Walter Roberson]

Yes, a model function is absolutely required.

It can be proven (I have posted constructed proofs in the past) that every finite dataset of finite precision has an uncountable infinity of models that fit it exactly (to within round-off error.)

With there being an uncountable infinity of models, the probability that any one of the models is the "right" model, is zero. There is no way for any kind of automated fitting system to tell the difference between the models... unless you have a model function that you can fit the parameters to.