Dear all,
I have a technical question regarding parallelization.
The following occurs:
The method was FOCEI, Simple ADVAN model
Without parallelization the NONMEM run ends successful without any error
messages.
After parallelization (8 nodes) the same run stops with the following error
message:
ON WORKER: WORKER1,ON DIRECTORY: worker1/
0PRED EXIT CODE = 1
OCCURS DURING SEARCH FOR ETA AT INITIAL VALUE, ETA=0
PK SUBROUTINE: ERROR IN COMPUTATION
ATTEMPT TO COMPUTE BASE**POWER WITH BASE<0.
(same for the following subjects)
>From a modeling perspective this is easy to solve (just constrain the
>respective THETA >0, it was coded as THETA**COV) but I ask myself why the
>THETA (=BASE) used as an initial by NONMEM was different comparing a single
>node run and parallelization.
Up to now I thought that the final calculation providing the "new THETA" is
done by the manager node -> in this case both runs should be identical, this
seems to be obviously not true.
Any information is highly appreciated.
Many thanks in advance
Dirk
Freundliche Grüße / Best regards,
Dirk Garmann
Senior M&S Expert
[cid:[email protected]]
Bayer Pharma Aktiengesellschaft
BPH-GDD-CS-CP-QP
42113 Wuppertal
Elberfeld, 0431
Tel.: +49 202 36 5577
Fax: +49 202 36 3998
E-Mail: [email protected]<mailto:[email protected]>
Web: http://www.bayer.com/
Vorstand: Andreas Fibig, Vorsitzender | Manfred Vehreschild, Hartmut Klusik
Vorsitzender des Aufsichtsrats: Richard Pott
Sitz der Gesellschaft: Berlin | Eintragung: Amtsgericht Charlottenburg HRB 283 B
<<inline: image001.gif>>