Dear Kajsa, I understand that if multiple mathematical relationship between a covariate and a parameter is included, some erroneous answers might be produced by the program. However, in this case, I am not exploring any mathematical problems. The major thinking behind this is to build on where it fails and test all that included afterwards. For some complex problems this might incur unreliable results, but regarding this model, although the structure is complicated but the relationship is pretty straightforward. It seems to me that I could have some luck out of this. To be more diligent, I would also use the method you mentioned above, and then compare the results just in case.
I am attaching the [code] section below. [code] *:BMI-2=PARCOV=(BMI/median)**THETA(1) *:RACE-2=PARCOV=1\ IF(RACE.EQ.2) PARCOV=THETA(1)\ IF(RACE.EQ.3) PARCOV=THETA(2) *:AGE-2=PARCOV=1\ IF(AGE.LT.12) PARCOV=THETA(1) *:SEX-2=PARCOV=(THETA(1)**(SEX-1)) *:STDY-2=PARCOV=1\ IF(STDY.EQ.22) PARCOV=THETA(1) Best Regards Xinting On 20 May 2014 16:16, Kajsa Harling <[email protected]> wrote: > Dear Xinting, > > The process you describe below will normally not give the results you > intend even if the scm program happens to run without error messages. If, > for example, in the new starting model you have WGT as a covariate on CL > with the linear relationship, and you have configured scm to try also the > exponential relationship for WGT on CL, then the procedure below will > erroneously give both the linear *and* the exponential relationship for WGT > on CL, instead of removing the linear relationship when trying the > exponential. For some special cases you might be lucky and not have errors, > but you cannot count on it. The procedure below is in conflict with how the > scm program was designed. > > Best regards, > Kajsa > > > On 05/20/2014 09:56 AM, Xinting Wang wrote: > > Dear Kajsa, > > Thanks very much for your reply. I found the answers you provided > cleared many unsolved problems. However, regarding below answer I found > there might be something different from the answer you provided. > > > *What if I use the unidentified relationships in the new run, and using a > mod file with relations in the scm logfile at the time of stop? * > > * That will usually not work.* > > While waiting for answers, I tested this method and found that it is > actually working. The procedure is as follows: > 1. use the included relations at the time of crash as the base model, > copy it from m1 folder to the folder where scm file exists [change $DATA > location]. > 2. make a copy of the original scm file, and rename it to a different > name. Some changes were made to the scm file: > - change the mod file to the new base model mod; > - delete all the included relations; > - change the search direction from both to forward. [A full backward > process would be initiated after the full model was established.] > 3. submit the scm file without the option of -directory. > > I am still waiting for the final results. But up until now it seems that > this method actually works. New covariate-parameter relations were added on > top of the previous ones. From my understanding this should be the same as > the method using [included_relations] as you mentioned, although this might > be a little bit more complex. > > Best Regards > > > On 20 May 2014 14:56, Kajsa Harling <[email protected]> wrote: > >> Dear Xinting, >> >> >> >> Previously when I wanted to resume a PsN run I used the -directory= in >> the command, and according to the PsN project homepage, this would enable a >> resume of a PsN run after stop. Such a method works fine without the >> -clean=3 option, but initiated a new search for covariates if I have >> -clean=3 option. However, this method seems to be different from the above >> mentioned. >> >> >> Yes, it is a different method. Setting -clean=3 removes files that are >> needed to resume an interupted run. I will make this more clear in the >> documentation, it is not explained properly at the moment. >> >> >> It seems to me that the method you mentioned requires 2 steps. The >> first one is to have only the included relations at the time of stop in the >> scm file. Should I use the -directory= option in the command line? >> >> >> The steps you need to perform are as follows: >> 1) Make a copy of your original scm configuration file, give it a new >> name and save it in the same folder as the original scm config file >> 2) open the original scm log file in the top level of the orginal scm run >> folder and locate the *last* place where it says "Relations included after >> this step", and copy those log-file-relations to the new configuration file >> under a new [included_relations] section at the end of the new >> configuration file. An example of a [included_relations] section with >> correct syntax is found in >> http://psn.sourceforge.net/pdfdocs/config_template_backward.scm and >> other details are found in the scm userguide. >> 3) If you had set option directory in your original config file then >> change the setting of that option in the new config file to something >> different >> 4) start a new scm with the same command as you used for the original >> scm, except that you use the new config file instead of the old, and either >> set -directory to the name of a folder that does not yet exist or omit the >> -directory option completely in which case the program will select a unique >> name for you. >> >> >> Secondly, if I start a new run in a new directory, why should I use the >> exactly same original input? >> >> You do not use exactly the same input, you use a new setting of >> [included_relations]. The input model however should be exactly the same. >> >> >> By doing so, would be program being able to skip the already >> established covariates and start from there onwards? >> >> >> Yes, it will start from the set of relations specified in >> [included_relations]. >> >> What if I use the unidentified relationships in the new run, and using >> a mod file with relations in the scm logfile at the time of stop? >> >> That will usually not work. >> >> >> >> One last small question is that a few days ago, an error message saying >> "psn.lst" is not created [something like this] was put forward by the >> program during the middle of searching. The cov-par relation was completed >> successfully before crush. Do you have any idea what might be the possible >> reasons of this? >> >> I need the complete error message in its context of the scm run >> messages to be able to answer your question. >> >> Best regards, >> Kajsa >> >> >> >> -- >> ----------------------------------------------------------------- >> Kajsa Harling, PhD >> System Developer >> Department of Pharmaceutical Biosciences >> Uppsala University >> [email protected]+46-(0)18-471 4308 >> http://www.farmbio.uu.se/research/researchgroups/pharmacometrics/ >> ----------------------------------------------------------------- >> >> > > > -- > Xinting > > > -- > ----------------------------------------------------------------- > Kajsa Harling, PhD > System Developer > Department of Pharmaceutical Biosciences > Uppsala University > [email protected]+46-(0)18-471 4308 > http://www.farmbio.uu.se/research/researchgroups/pharmacometrics/ > ----------------------------------------------------------------- > > -- Xinting
