Dear allstat,
We have a problem with the SAS/IML non linear opimization procedures such as
NLPTR, NLPNRA, NLPQN and NLPCG. Let's call them the "NLOP"
We are trying to fit a gompertz survival distribution on simulated survival
data (obtained from piecewise exponentials + censoring). We are first
computing the starting values by non-linear regression (NLIN). These values
are then used as starting values in the log likelihood non linear
optimization procedure. Everything goes well for the first 50-200
simulations. Then the "NLOP" (at least those I have tested, see above)
generates a fatal error with the following message attached:
ERROR: The function value of the objective function cannot be computed
during the optimization
process.
We thought that providing the analytical form of the gradient/hessian would
solve but without success.
What astonished us is that for errors like:
WARNING: Optimization routine cannot improve the function value.
ERROR: TRUREG Optimization cannot be completed.
the procedure resumes and the next simulation is entered, documenting the
previous one with a return code equal to -5 which can be detected and
treated appropriately.
This is what we would expect for all errors. On the contrary, after the
fatal error has occured, the "NLOP" are returning the same output for all
the next simulated datasets and no specific return code is output.
The RESUME statement has no effect. What is probably needed is a
reinitialization of the the "NLOP" but how to do that ? Relaunching the PROC
IML procedure ?
Has somebody an idea ? Any suggestion is welcome.
Thierry Gorlia
----------------------------------------------------------------------------
----------------
Thierry Gorlia Email : [log in to unmask]
Statistician
Health Economic Unit
Quality of Life Unit
European Organization for research and Treatment of Cancer
Data Centre - Avenue Mounier 83, 1200 Brussels, Belgium
Phone : +32 2 774 16 80
Fax : +32 2 779 45 68
Web site: http://www.eortc.be/home/qol/
----------------------------------------------------------------------------
----------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|