cvode, odeSD, etc.
Sebastian Weber
I just saw that the cvode branch is MERGED in stan-math & stan!!! Great news! I am looking forward to test this.
I am currently working jointly with the odeSD author to get this stiff integrator into Stan. Would that still be considered? I.e. the integrator promises
- a ~5x speedup over cvode (stiff system)
- is implemented as header only C++
- as Eigen's linear algebra is used which takes advantage of the MKL, those people having it may get another ~2x speedup
The down-side is that odeSD requires second order derivatives and hence hessians must be calculated.
Thoughts on this?
A first (somewhat restricted in functionality) prototype version of this in Stan is doable this month.
Cheers,
Sebastian
PS1: Really, great work! I am glad this happened so fast now.
PS2: Would you consider to let users choose between the non-stiff and stiff cvode integrator? Or is the non-stiff integrator simply the boost integrator?