How to avoid negative intermediate guess in nlmefit

Hello everyone,

when I use nlmefit, it always comes up with some negative guess after a few iterations, these negative values crash my ODE solver. Is there a way to avoid this?

The easiest way I can think of is to set lower bound for fitted parameters, however, I cannot figure out how to do that with nlmefit.

Thank you!

Rui

Answer 1

[Tom Lane]

If a parameter P needs to be positive, sometimes people will instead estimate a parameter logP and then use exp(logP) in place of P in their function.

[Rui]

Hi Tom,

Thanks! I saw there's an option "ParamTransform", should I use this option, or should I explicitly take logP? Are these two equivalent?

If I estimate logP and use exp(logP) in my model, would that affect the estimation of covariance matrix? I mean should I somehow covert it back to normal space using exp()?

Rui

[Tom Lane]

I'd forgotten about this parameter. Yes, I'd use that to specify a log transform and let the function take care of converting the covb matrix.

[Rui]

Thank you!