What is the fastest way to estimate many models with different initial conditions and iter.max = 0?

[billdenney]

In the near future, I'd like to implement sampling-importance-resampling (SIR). The way that this is done within PsN is that it creates thousands of $EST steps within a single NONMEM control stream. Is something similar feasible within nlmixr2est::nlmixr2()?

I think that the answer is no, and so the fastest way that I can think to do it is many calls to nlmixr2, but I would guess that there is some step deeper in the estimation process where the model file has been processed and possibly just before a C++ call where we may be able to provide new initial estimates. I would also guess that part is not exported.

[mattfidler]

This needs to be probably discussed in more depth (probably at a developer meeting). There are many way to speed this up, though I need to understand the problem a bit better to be able to help here.