--
You received this message because you are subscribed to the Google Groups "stan development mailing list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stan-dev+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
<metrics.pdf>
I now see from later email that it's two different solvers
built into CVODES. Do we have:
stiff: CVODES > CVODE
non-stiff: CVODES > Boost
I like the idea of the 4 proposed functions, i.e. integrate_ode/integrate_ode_stiff with/without control of abs+rel tolerances + maximum steps.
It feels bad to ditch RK45 from boost... I will try to benchmark Adams vs RK45 on a few cases to get an opinion on that, but probably you are right in that boost odeint is not needed anymore.
In what sense do you want to maintain back-ward compatibility? I mean the integrate_ode_cvode function only ever existed on the develop branch. This is why I was thinking it would be ok to change it even now.
@Michael: I think Bob meant that CVODES has a built-in non-stiff and a stiff solver. I agree that the terminology is confusing. We have the CVODES solver suite which includes Newton (stiff) and Adams Moulton (non-stiff).
You both were too fast for me in posting (and I am too tired). We are cross-posting.
I can live with that proposal too. Still, I would like to understand at what point you guys consider that we need to maintain back-wards compatibility...
On Wednesday, April 6, 2016 at 12:10:33 AM UTC+2, Bob Carpenter wrote:
> See inline.
>
> > On Apr 5, 2016, at 5:26 PM, Sebastian Weber
>
- Bob
However, I just recalled, that the function got official mentioning on the blog:
http://andrewgelman.com/2015/12/16/working-stiff/
But I guess this does not change the fact that we can dump it now or does the blog make the function worthy to maintain?
Sebastian
On Wednesday, April 6, 2016 at 1:27:33 AM UTC+2, Bob Carpenter wrote:
> Great. No issue removing something we haven't released.
> I don't want to get into the business of supporting snapshots
> of develop!
>
> - Bob
>
> > On Apr 5, 2016, at 7:25 PM, Daniel Lee
> >