Hi Paolo and Rob,
I’ve been experiencing similarly large CWRES (no “I”) values with NM7.4.1 for
models which produced reasonable values when run unchanged in NM7.3. Other
colleagues also saw this behavior.
I believe we typically see it with bigger/more complex models (can’t recall a
case where we had the problem with an analytical ADVAN).
Since computation of CWRES is parallelized since 7.4.1, so I guess something
about the implementation may have been changed to enable this.
I haven’t tried it myself, but if the “old” code is still used when the
computation is not parallelized, one could speculate that disabling parallel
evaluation of the residuals might help.
The one thing I tried that did not help was using WRESCHOL.
Kind Regards,
Sven
Von: [email protected] [mailto:[email protected]] Im
Auftrag von Bauer, Robert
Gesendet: Donnerstag, 16. November 2017 19:46
An: [email protected]
Betreff: [NMusers] Extreme values of CWRESI in NONMEM 7.4.1?
Paolo:
I reviewed my NONMEM 7.4 testing results on CWRESI, and they provide the same
values in NONMEM 7.3 as well as NONMEM 7.4. In case you ran the problem in
parallel, because NONMEM 7.4 now executes the diagnostic evaluation in parallel
as well, I tested under parallel execution conditions, and the CWRESI values
are also the same between NONMEM 7.4 (parallel) and NONMEM 7.3. Incidentally,
as a diagnostic that is expected to be normally distributed when the model is
correct, CWRES is preferred even when INTERACTION is used in the $EST step.
I would be happy to run your problem and see if I can detect something more
subtle that is involved in your case.
Robert J. Bauer, Ph.D.
Senior Director
Pharmacometrics R&D
ICON Early Phase
820 W. Diamond Avenue
Suite 100
Gaithersburg, MD 20878
Office: (215) 616-6428
Mobile: (925) 286-0769
[email protected]<mailto:[email protected]>
http://www.iconplc.com/