RE: Simulation problem of normal distributed data

From: Erik Olofsen Date: July 30, 2014 technical Source: mail-archive.com
Hi Gerry, EXP(-X) might become such a small number that PRN no longer increases. For example, 0.5+S is represented as 0.5 for small S. Using verbatim code (see nmhelp) you could print N, PY and PRN to see if something like this occurs. Best regards, Erik
Quoted reply history
________________________________ From: [email protected] [[email protected]] on behalf of Gerry Sheng [[email protected]] Sent: Wednesday, July 30, 2014 10:40 AM Cc: nmusers Subject: Re: [NMusers] Simulation problem of normal distributed data Dear Goyal, Thank you for your advice. I changed the second seed to 17545, but the new cutoff number of records is 1498 . If data records>1498, the same problem still exists. Best wishes, Yucheng Sheng Research Associate School of Pharmacy, UCL On 29 July 2014 23:25, Navin Goyal <[email protected]<mailto:[email protected]>> wrote: Hi Sheng, I don't know the exact reason behind such mysterious behavior but if you just change the seed number (the UNIFORM seed) it works. e.g. I changed the seed# from 7545 to 17545 and was able to simulate 1000 rows for the subject. $SIMULATION (6446) (17545 UNIFORM) ONLYSIM NOPRED SUB=1 Its a rare occurrence but may be the seed number caused some issue (while generating sufficient samples to satisfy the constraints you set up using the DOWHILE LOOP). For example I think the same seed number (7545) may work when the DOWHILE condition is reversed (RL .LT. PRN) however it may not be appropriate based on your problem. .......or may be there is some other explanation to such occurrence. Kind regards Navin Goyal On Tue, Jul 29, 2014 at 10:54 AM, Gerry Sheng <[email protected]<mailto:[email protected]>> wrote: Dear NMuser, I encountered a strange problem when I used the following code to simulate normal distributed data. When the data records >797, no result appears with NONMEM run into an infinite loop . But when the records decreasing to 796 (delete the last row), the results will occurred immediately. Anyone else met this phenomenon? Any comments and suggestions are appreciated. Thank you in advance. $PROBLEM PD $INPUT ID,CONC,DV $DATA data.csv IGNORE=@ ; $PRED NMEAN=THETA(1)+ETA(1) SD=THETA(2) ;Simulate count IF (ICALL.EQ.4) THEN N=0 PRN=0 CALL RANDOM(2, R) RL=1-R DO WHILE (RL .GT. PRN) LNNOR=LOG(6.283185)+2*LOG(SD)+((N-NMEAN)/SD)**2 PY=EXP(-0.5*LNNOR) PRN=PRN+PY IF (PRN .LT. RL) N=N+1 ENDDO DV=N ENDIF $THETA (30,41.7,60) ;MEAN (2,14.4,50) ;SD $OMEGA 0 FIX ;MEAN $SIMULATION (6446) (7545 UNIFORM) ONLYSIM NOPRED SUB=1 $TABLE ID TIME DV ONEHEADER NOPRINT FILE=npctab1.dta data.csv like this ID CONC DV 1 0 . 1 0 . 1 0 . 1 0 . 1 0 . 1 0.01 . 1 0.01 . 1 0.01 . 1 0.01 . 1 0.01 . ..... Yucheng Sheng Research Associate School of Pharmacy, UCL -- Navin Goyal
Jul 29, 2014 Gerry Sheng Simulation problem of normal distributed data
Jul 29, 2014 Navin Goyal Re: Simulation problem of normal distributed data
Jul 29, 2014 Robert Bauer RE: Simulation problem of normal distributed data
Jul 30, 2014 Gerry Sheng Re: Simulation problem of normal distributed data
Jul 30, 2014 Erik Olofsen RE: Simulation problem of normal distributed data