In NONMEM IV as distributed, the number of data records in a single
individual record must not exceed 400. I wonder if the additional data
records sometimes cause this restriction to be exceeded. NONMEM IV
does not detect violations of this restriction, and might crash with an
I/O error of some sort.
Dr. Harnisch should examine his data set. If there are more than 400
data records in some individual records, he can remove the excess the
additional records. Alternately, NONMEM source code can be changed to
allow more records. This is described in Users Guide III, 3.2.9,
"Changing the Size of NONMEM Buffers". Buffers 1 and 2 must be
changed.
He says he adds records such as:
EVID=0, PCMT=CMT=2, CONC=DV=0, and MDV=1
or
EVID=2, PCMT=CMT=2, CONC=DV=0, and MDV=1
Ruedi Port was basically correct when he responded:
> (To me, EVID=2 seems to be the more correct version for the added
> lines with MDV=1.)
EVID=2 is more in the spirit of how things should be done, but in fact
the effect on the run itself would be insignificant. (There is a
difference if the $ERROR block specifies CALLFL=0, but few people
use this feature.)
Ruedi is also correct in advising Dr. Harnisch to omit the $ESTIM
record when the only purpose of the run is to obtain predictions
at additional data points. This will take care of the inability
of the search to proceed because it is already at the miniumum.
The error message from PREDPP is more puzzling:
NUMERICAL DIFFICULTIES WITH INTEGRATION ROUTINE.
MAXIMUM NO. OF EVALUATIONS OF DIFFERENTIAL EQUATIONS, 100000, EXCEEDED.
0PROGRAM TERMINATED BY PRRES
This message might legitmately occur with a very large data set. It is
possible to raise the limit, but first it would be best to make sure
that this is not some side-effect of the 400-record problem.
If it persists, Dr. Harnisch should contact me and I'll tell him what
to do.
Prediction Curve
did you try to skip the estimation step when running your MSFI file (running
the $TABLE step only)?
(To me, EVID=2 seems to be the more correct version for the added lines
with MDV=1.)
Hi. We are running NONMEM on HP-UX 9.03 for a long period. Using
a nonlinear model (ADVAN8 including linear and nonlinear
elimination processes and different terms for inter occasional
variabilities), I had the wish to make predictions on all non
measured time points.
Sometimes it is working using these additional records:
EVID=0, PCMT=CMT=2, CONC=DV=0, and MDV=1
or
EVID=2, PCMT=CMT=2, CONC=DV=0, and MDV=1
and a MSFI-file which is the result of a prior succesfully
finished run without these add. records and a MSFO-file.
But in most cases NONMEM crashes with the following
*** FORTRAN I/O ERROR 933: END (OR BEGIN) OF FILE WITH NO "END=" SPECIFIER
FILE: FILE19, UNIT: 19
LAST FORMAT: (34(1X,1P,E11.4))
( 0) 0x80fbb06c io_ded + 0x31c [/usr/lib/pa1.1/libcl.sl]
( 1) 0x80fbea9c read_seq_unform + 0x1c8 [/usr/lib/pa1.1/libcl.sl]
( 2) 0x80fbec68 new_rc + 0x90 [/usr/lib/pa1.1/libcl.sl]
( 3) 0x80fcf9cc FTN_S_RSUE + 0x158 [/usr/lib/pa1.1/libcl.sl]
( 4) 0x0001f984 dat6 + 0x5d4 [./nmd.27205]
( 5) 0x00039bd8 outpt + 0x25e8 [./nmd.27205]
( 6) 0x00013fc4 final + 0x494 [./nmd.27205]
( 7) 0x000102b4 nonmem + 0x22c [./nmd.27205]
and in the output
0PRED EXIT CODE = 1
0INDIVIDUAL NO. 1 ID= .11200000E+03 (WITHIN-INDIVIDUAL) DATA REC NO. 91
THETA=
2.83E+00 5.17E+00 8.87E+00 2.54E+01 9.48E-01 3.39E+00 5.70E-01
5.80E-03 1.01E+01
NUMERICAL DIFFICULTIES WITH INTEGRATION ROUTINE.
MAXIMUM NO. OF EVALUATIONS OF DIFFERENTIAL EQUATIONS, 100000, EXCEEDED.
0PROGRAM TERMINATED BY PRRES
Excluding all the additional points, the run is coming to a
successfully end using the same MSFI-file.
In both cases NONMEM terminates with
0MINIMIZATION TERMINATED
DUE TO INABILITY TO PROCEED BEYOND TERMINATION POINT OF PREVIOUS SEARCH
NO. OF FUNCTION EVALUATIONS USED: 0
NO. OF SIG. DIGITS IN FINAL EST.: .0
but this is due to the nonchanging SIGDIGITS from the first to the
second run, and changing this number is only increasing the CPU-time, which
is for the first run about 4 hours (only FO and POSTHOC) !!
Did I make some mistakes in including the additionals ?
I have had similar problems with HPUX 9.01. I could never find the cause
of these mystery crashes due to the file with no "END=" error. I usually
worked around the problem by some other change to the model. It would be
interesting to try your problem on a different CPU. Why dont you ask
Alison Boechmann if she will run it for you and we can find out if it is
a HP-UX specific error?