Re: Error in LSODA: Code -5
Setting lower TOL values (as low as 2-3-4) may help as well (but it should be done carefully as there are examples where TOL=4 solution was different from TOL=9 solution; so if this helps, TOL should be increased back to 7-8-9 later in the modeling process when the more stable model is found)
Leonid
--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566
Quoted reply history
On 12/11/2015 9:40 AM, [email protected] wrote:
> Usually tolerance criteria for ODE solvers apply to the solution at the end of an integration
> interval. In that case, there is a difference between the solution at some time instant if it is
> preceded by an observation record with MDV=1 or not. So the tolerance settings for ADVAN13 which
> are ok for the "normal" subjects could be problematic for the "deficient"
> subjects if an integration interval gets too large. So perhaps adding missing observation records
> in between could help preventing LSODA errors.
>
> Erik
> ________________________________________
> From: [email protected] [[email protected]] on behalf of
> Andre Jackson [[email protected]]
> Sent: Thursday, December 10, 2015 9:22 PM
> To: 'Leonid Gibiansky'; [email protected]
> Subject: RE: [NMusers] Error in LSODA: Code -5
>
> Leonid:
>
> I tried the first suggestion by getting rid of some of the data that was
> deficient (i.e., subjects with less than 8 concentrations over the 30 hr.
> sampling period) and the LSODA error was not present.
>
> I will also try your other suggestions sequentially to see if they also
> improve the analysis.
>
> Thanks for your input.
>
> Andre
>
> -----Original Message-----
> From: Leonid Gibiansky [mailto:[email protected]]
> Sent: Thursday, December 10, 2015 9:59 AM
> To: Andre Jackson; [email protected]
> Subject: Re: [NMusers] Error in LSODA: Code -5
>
> Usually this error message is not informative: it just tells you that
> something is wrong with the model. It can be fixed by (in the order of
> importance)
> - cleaning the data (look on concentration-time plots to remove obvious
> outliers),
> - starting with the better initial parameter values (e.g., your mean
> analysis final estimates),
> - setting bounds on THETAs,
> - removing some ETAs,
> - using combined residual error instead of the purely proportional one,
> - using residual error with ETAs (instead of Y=IPRED*(1+W*EPS()) use
> Y=IPRED(1+W*EXP(ETA())*EPS()),
> - other model modifications/simplifications/improvements, including changing
> the estimation method (try FO if nothing helps, and then used FO final
> estimates to start something more advanced)
>
> Leonid
>
> --------------------------------------
> Leonid Gibiansky, Ph.D.
> President, QuantPharm LLC
> web: www.quantpharm.com
> e-mail: LGibiansky at quantpharm.com
> tel: (301) 767 5566
>
> On 12/10/2015 8:10 AM, Andre Jackson wrote:
>
> > I have run a control stream that runs okay following MU modeling with
> > the mean data to establish the best structural model using NM7 with
> > ADVAN 13 TRANS1 TOL=6.
> >
> > However when I attempt to run the model in a population mode with even
> > one eta on a MU I get the following error.
> >
> > OCCURS DURING SEARCH FOR ETA AT INITIAL VALUE, ETA=0
> >
> > ERROR IN LSODA: CODE -5
> >
> > ERROR OCURRED WHILE ATTEMPTING TO OBTAIN INITIAL VALUES FOR ATOL
> >
> > 0PROGRAM TERMINATED BY OBJ
> >
> > I have seen some posts in the archives on LSODA and differential
> > equations but nothing that seems to address my problem directly.
> >
> > Does anyone know of any more definitive posts or have suggestions as to
> > how best to proceed.
> >
> > Thanks
> >
> > Andre Jackson
> >
> > [email protected] <mailto:[email protected]>