You are here

Using analyticaly derived gradient and hessian in OpenMX

2 posts / 0 new
Last post
Anonymous's picture
Anonymous (not verified)
Using analyticaly derived gradient and hessian in OpenMX

I know NPSOL will accept first derivatives (gradient) and second derivatives (Hessian) of the fitting function with respect to the parameters.

If I have determined these analytically for my model, can i make OpenMX parse them to NPSOL?

The only slot I can find for the gradient and the hessian are in a fitted model object not in an model hat still has to be fitted.

I did find the mxOption slot where i can tell NPSOL the gradient and hessian will be provided and need not be estimated


jpritikin's picture
Joined: 05/24/2012 - 00:35
choice of optimizers

In the next version of OpenMx, you will have a choice of optimizers. One of the choices is a simple Newton-Raphson optimizer that is faster than NPSOL but requires an analytic gradient and hessian. Currently, this is only available in source code control. We plan to release a beta of this code soon.