NONMEM Users Network Archive

Hosted by Cognigen

RE: $COV UNCONDITIONAL PRINT=E

From: Mark Sale - Next Level Solutions <mark>
Date: Mon, 26 Oct 2009 09:17:12 -0700
Sam,
   Many years ago Stuart scold= ed me for using the Unconditional option, saying that a model that didn't c= onverge had no chance of a successful covariance step (I didn't ask why he = put the option in if he didn't want people using it).  But, I have had= put a couple of models that appeared to be "hung" into the development env= ironment and found that they were not, in fact hung.  Almost invariabl= y, the "hung" part is in the call to OBETA (I assume you're using FOCE?), O= BETA is a minimization in itself, and while the main NONMEM minimization ha= s a limit (MAXEVAL) OBETA doesn't seem to.  So, if the minimization (i= n OBETA) is difficult it will just keep running.  Options are to belei= ve those who claim covariance doesn't mean anything, or wait for the parall= el version of NONMEM.

Mark

Mark Sale MD
Next Level Solutions, LLC
www.NextLevelSolns.com
91= 9-846-9185

-------- Original Message --------
Subject: [NMusers] $COV UNCONDITIONAL PRINT=E
From: Sam Liao <sliao Date: Mon, October 26, 2009 11:36 am
To: <nonmem
Dear NONMEM team:
I have two nm7 questions.
<= span style="font-size: 11pt; font-family: "Calibri","sans-= serif"; color: rgb(31, 73, 125);"> 
= I have a PKPD model using ADVAN13 t= o solve the ODE.  It took 12 hours to complete the $EST step with roun= ding error.
But it’s been over 15 hours running the $COV step with UNCOND= ITIONAL option.
 
This is much longer time than I expected.  Could this be a= n infinitive loop?  Should I interrupt the run and try smaller  S= IGL in $COV?  I tried the CTL-K or CTL-E to exit, but it did not work = for me.
 
The second question related to BAYES method in $EST step.  The mod= el run successfully using FOCE method.  But when I tried the BAYES met= hod with NOABORT option, it terminated with ‘OBJECTIVE FUNCTION IS = INFINITE’ in Burn-in mode after 10 iterations.  The statement = used as below.  How can I get around this problem?
========= ===============
= $SUBROUTINES ADVAN13 TOL=6 &= nbsp;           &nbs= p;            &= nbsp;           &nbs= p;            &= nbsp; 
$EST METHOD=BAYES NBURN=100 NSAMPLE=3000 SIGL=6 NOABORT FILE= =RUN136.TXT
 
Best regards,
Sam
 
 
Pavel:
The objective functio= n progress looks good.  You should expect some Monte Carlo fluctuation= s.  You should also run more iterations (perhaps NITER=200), and set= CTYPE=3, which turns on the termination tester.  To resume where yo= u left off, rename your new control stream file, and put in the following l= ines.
 
$EST METHOD=CHAIN NSAMPLE=0 ISAMPLE=50 FILE=my_old_c= ontrol_stream_file.ext
$EST METHOD=IMP NITER=200 CTYPE=3 FILE=my_new_control_stream_fil= e.ext
 
Make sure you are linear MU referencing to get the greatest = efficiency.
 
 
 
 = ;

=

From: owner-nmusers ="_blank" onclick="if(window.location==top.location){Popup.composeW= indow('pcompose.php?sendto=owner-nmusers%40globomaxnm.com');}else{top.Pop= up.composeWindow('pcompose.php?sendto=owner-nmusers%40globomaxnm.com');};= return false;" href="mailto:owner-nmusers musers Sent: Saturday, October 24, 2009 8:03 PM
To: nmusers xnm.com
Subject: [NMusers] method=ITS, OPTIMIZATION NOT TESTED= (?!)

Hello NONMEM Team,
I= found method=imp useful when there are local maxima.  Nevertheless,= at the end of optimization it prints a message, which makes me feel somewh= at uncomfortable:  OPTIMIZATION NOT TESTED.  Also, the final obje= ctive function is not always the lowest one.  An example is below.&nbs= p; How do we interpret the results in this case? 
 THETAS THAT ARE SIGMA-LIKE:
 MONITORING OF SEARCH:
 iteration       &n= bsp;    0 OBJ=    10625.663135214874 &= nbsp;  
 iteration      &= nbsp;     1 OBJ=    10601.188754983375=     
 iteration     =        2 OBJ=    10537.89511= 4114934    
 iteration    = ;        3 OBJ=    1047= 1.674625518765    
 iteration  &nbs= p;         4 OBJ=  &nbs= p; 10430.297437731866    
 iteration &nb= sp;          5 OBJ= &nb= sp;  10461.973668565577    
 iteration&n= bsp;           6 OBJ=&n= bsp;   10462.638834406265    
 iter= ation            7 O= BJ=    10423.464983371641    
&nb= sp;iteration          &nb= sp; 8 OBJ=    10417.959956991735     <= br>  iteration         &n= bsp;  9 OBJ=    10417.594007447198   &= nbsp;
 iteration        &= nbsp;  10 OBJ=    10414.708468642830   = ; 
 iteration        = ;   11 OBJ=    10427.810693855947  &nb= sp; 
 iteration       &nb= sp;   12 OBJ=    10412.889081059604  &= nbsp; 
 iteration       &= nbsp;   13 OBJ=    10411.980622268416  = ;  
 iteration       = ;    14 OBJ=    10424.501127174915 &nb= sp;  
 iteration      &nb= sp;    15 OBJ=    10416.332869468861 &= nbsp;  
 iteration      &= nbsp;    16 OBJ=    10416.622580251338 = ;   
 iteration      = ;     17 OBJ=    10412.401585537709&nb= sp;   
 iteration     &nb= sp;     18 OBJ=    10415.117257355550&= nbsp;   
 iteration     &= nbsp;     19 OBJ=    10415.30237096105= 5    
 iteration     = ;      20 OBJ=    10409.066188189= 252    
 iteration    &nb= sp;      21 OBJ=    10413.7806204= 68329    
 iteration    &= nbsp;      22 OBJ=    10410.78749= 6174480    
 iteration    = ;       23 OBJ=    10410.633= 582415931    
 iteration   &nb= sp;       24 OBJ=    10409.9= 70257443048    
 iteration   &= nbsp;       25 OBJ=    10409= .702420124611    
 iteration   = ;        26 OBJ=    104= 09.213115058612    
 iteration  &nb= sp;        27 OBJ=    1= 0409.690639357370    
 iteration  &= nbsp;        28 OBJ=   = 10410.016047785200    
 iteration  = ;         29 OBJ=  &nbs= p; 10408.157468814226    
 iteration &nb= sp;         30 OBJ=  &n= bsp; 10407.779614704938    
 iteration &= nbsp;         31 OBJ=  =   10410.164563157052    
 iteration = ;          32 OBJ= &nbs= p;  10408.364552302961    
 iteration&nb= sp;          33 OBJ= &n= bsp;  10407.431920727997    
 iteration&= nbsp;          34 OBJ= =    10408.286189641487    
 iteratio= n           35 OBJ=&nbs= p;   10407.907347050501    
 iterat= ion           36 OBJ=&n= bsp;   10407.451608770069    
 iter= ation           37 OBJ==     10407.189482360372    
 it= eration           38 OBJ= =    10406.484357336147    
 = ;iteration           39 O= BJ=    10409.167125968375    
&nb= sp;iteration           40= OBJ=    10406.840873883246    
&= nbsp;iteration           = 41 OBJ=    10407.679485561714    
=  iteration          = ; 42 OBJ=    10405.341101045238      iteration         &nb= sp; 43 OBJ=    10404.704382334516     =
 iteration         &= nbsp; 44 OBJ=    10405.348023082915    = ;
 iteration         = ;  45 OBJ=    10405.347406984720   &nb= sp;
 iteration        &nb= sp;  46 OBJ=    10401.873473651774   &= nbsp;
 iteration        &= nbsp;  47 OBJ=    10404.036204419035   = ; 
 iteration        = ;   48 OBJ=    10405.072916975221  &nb= sp; 
 iteration       &nb= sp;   49 OBJ=    10402.976628923887  &= nbsp; 
 Elapsed estimation time in seconds: 30420.73
&nb= sp;iteration           50= OBJ=    10403.285958168881     <= /o:p>
 #TERM:
 OPTIMIZATION NOT TESTED
Thanks,=
Pavel
 
 =
 
 
=
ICON plc made the following annotations.
-------=
-----------------------------------------------------------------------
This e-mail transmission may contain confidential or lega=
lly privileged information
that is intended only for t=
he individual or entity named in the e-mail address. If you
are not the intended recipient, you are hereby notified that any disc=
losure, copying,
distribution, or reliance upon the co=
ntents of this e-mail is strictly prohibited. If
you h=
ave received this e-mail transmission in error, please reply to the sender,=
 so that
ICON plc can arrange for proper delivery, and=
 then please delete the message.
Thank You,=
ICON plc
South County Business Park
Leopardstown
Dublin 18
<= pre>Ireland
Registered number: 145835
=
 

left.letterhead
(image/png attachment: left.letterhead)

Received on Mon Oct 26 2009 - 12:17:12 EDT

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to: nmusers-request@iconplc.com.

Once subscribed, you may contribute to the discussion by emailing: nmusers@globomaxnm.com.