ODE branch status?

[Bill Harris]

You're having such a good and important conversation about new features that I almost hate to intrude with a specific question, but I will.

Is there any news about the status of the ODE development work? I may have a problem where I'd like to use such a tool very soon (and I may not). If I do, I see at least three choices:

- the ODE features that are coming in Stan

- deSolve and FME in R

- MCSim

If Stan's ODE capabilities are good enough to use and buildable by a non-expert today, I might be tempted to try it (a bit risky for me, as this is for real work, not "simply" research). If it's not that close, I'd seek out another alternative this time around.

Thanks,

Bill

[Bob Carpenter]

We won't actually have ODE in Stan's modeling language
for a while.

You can follow the integrator branch to see how we implemented
an example using Boost's integrators. We're working on some
other models in a private branch, and those have also worked
well.

But it's going to require C++ coding of the solver as we
did in the integration branch and then building the resulting
function into Stan. It's not that much work, but it's completely
undocumented so far.

And there's no support for stiff equations yet.

I have no firsthand experience with deSolve/FME in R
or with MCSim. We've been consulting with Frederic Bois,
the developer of MCSim, on the work for Stan, and we've
used some code he's written in deSolve to validate the
integrator in Stan. But I haven't actually used MCSim
or deSolve myself.

- Bob

[Bill Harris]

Bob Carpenter <ca...@alias-i.com> writes:

> We won't actually have ODE in Stan's modeling language
> for a while.

Thanks, Bob. I figured as much, but I didn't want to start down one
path Monday morning and read about the release of that branch Monday
evening. :-) Good luck with it.

> I have no firsthand experience with deSolve/FME in R
> or with MCSim. We've been consulting with Frederic Bois,
> the developer of MCSim, on the work for Stan, and we've
> used some code he's written in deSolve to validate the
> integrator in Stan. But I haven't actually used MCSim
> or deSolve myself.

I've not used deSolve/FME yet, either, except to run a toy case.

I rather like the expressiveness of MCSim (and it promises to be even
more expressive, now that it can call deSolve). There are some short
examples in its documentation and one in the quick reference guide on
http://www.gnu.org/software/mcsim/. My only real problem with MCSim is
that I've lost the recipe for building it on Windows, and I do sometimes
need a Windows version.

Bill
--
Bill Harris
Facilitated Systems
http://makingsense.facilitatedsystems.com/