Hi Pharsight, I don’t really expect you’d know this but wondering whether you think I should stop my run and restart, given its been running for about 20 hours and hasn’t completed an interation in about 8 hours (i.e. last iteration was 8 hours ago and its been stuck on the 26th iteration since then). I"ve got 117 subjects and about 2000 datapoints so its a lot of data and I know these should take a while but could it be NLME is just stuck? Before you ask my computer is not very powerful. Its a duo core with 2.5 GHz but when it was installed it wasn’t installed to take advantage of MPI which I will be rectifying. I’m running ELS algorithm. Any thoughts would be appreciated. Elliot
Dear Elliot If we are talking about differential equations, “stuck” issue can happen and I experienced that too. I will convey your concern to the development team to see if there is way to know the status of the integration procedure. best Regards; Serge
Hi Serge, Yes, it appears its an integration step or whatever its doing when the application is in the iterative phase. Elliot
Can you try the QRPEM engine option. Select 1000 for the number of samples, 100 for the max number of iterations and click on advanced option and select the MAP feature. Let me know how it proceeds. Now if you kill the process, please go to task options (ctrl/alt/delete) and in star6t task manager, go to process and kill the mpi process (the first one you see). If you are not using mpi, kill the nlme7.exe process (if it is there). Sometime, these processes are not stopped and rerunning causes problems. Best Serge
Well I can tell you its running and in 4 minutes its done 3 iterations. That’s better than the ELS which did about 1 an hour 
I’ll post tomorrow to let you know how it goes. I was relying on ELS because I could relate it a NONMEM algorithm and had done some of my related work using ELS. Can you confirm if the results should be the same for QRPEM vs. ELS? I think I read that in the user manual and the advantage is the processing speed but I’d hate to be comparing AIC and theta’s for example using two different algorithms if they weren’t pretty similar. Thanks Elliot
Dear Elliot Look at the -2LL. If it goes down and converge, then you are in good shape. When QRPEM is working, then it is an unbiased optimization algorithm. The problem is that you need to learn a little bit about how to use that engine and also the QRPEM will be way better in the next release (few months I think). If you are using the QRPEM, start always with big variances (0.3 at least in the log-domain). If you see the -2LL start to oscillate too much and/or GOES UP INSTEAD OF DOWN, then incrase the number of samples (1000 should work usually). Use always MAP advacned option. At each iteration, a MAP optimizatiojn is performed on each individual. If it does not work, write 10 as the number of burn in. This helps finding the individual posterior distributions. If it still does not work, then either you have bad initial estimates (use the initial estimate tab to fine tune your initial estimates), either the data have some individual posteriors that have too small variances and it is better to shift to ELS. Finally, try to run your model with a test dataa set that has way less number of individuals, let say 10%. This will give you a good idea about the time you should need for your big data set. I hope these tips will help you. best Serge