Speed up sampling of forced ODE model

[wds15]

Hi!

You tried the stiff solver already?

Currently you are using the defaults for the tolerances, so have a look at my post here:

ODE solver in PBPK model takes too much time

Hi! Linas is referring to a technique wrt to the analytic Jacobians which is rather advanced and requires you to write C++ code… not quite for starters. A few comments: You tolerances are way too large! Such imprecise ODE solutions lead to a collapse of the numerical precision. In popPK problems I found that a relative tolerance of 1E-9 (or 1E-7 is also OK depending on the case) and an absolute tolerance of 1E-5 tend to work good for the problems I have come across. The odd thing is when yo…

The default tolerances are quite conservative. So it is important to adapt the absolute tolerance to a meaningful value (the relative tolerance may also be increased, but from experience to a lesser extent).

Also, if I see that right, then you can just vectorize your for loop for the likelihood. If you write normal_lpdf(I | I_int, sigma) you get exactly the sum which you are forming afterwards.

I hope this helps.

Sebastian