[SciPy-user] PyDSTool

[Robert Clewley]

Alan, to answer your questions:

1. I don't think there's any such destiny :) ... inasmuch as we haven't 
presumed that our libraries would be considered either useful or 
well-written enough to be brought into the SciPy fold. Our classes are 
still effectively prototypes, and I would not like to get roasted for 
trying to integrate them into SciPy without a lot more time (and probably 
outside help) to get them up to scratch.

Now that we have shown proof-of-concept with some of our constructs, such 
as a Pointset, and some of the symbolic expression stuff, I expect that 
these could be broadly useful to scientific users in general. But I 
personally don't have much time to be a coder any more. If someone wanted 
to help me and/or give me feedback on what might be of use to SciPy users 
then perhaps I'll be less hesitant.


2. In short, no. It was a quick fix when we were trying to get any 
available integrator running with our core libraries, in order to test out 
our ideas for wrapped integrators in our simulation environment.

Other Python developers here at Cornell (Chris Myers, Jim Sethna, Ryan 
Gutenkunst, and others of the SloppyCell group) have used and adapted 
SciPy's LSODE solver in their package, and we have discussed the relative 
merits of trying to improve SciPy's support of dynamical systems through 
our "fancier" ODE solver interfaces.

Apart from a lack of time on our part to invest in maintaining 
publicly-releasable code, the other issue with PyDSTool is that users 
would have to invest in our basic classes to work with our wrapped 
integrators. Even though the UI permits very easy specification of ODE 
right-hand sides and other options, I am nevertheless worried that 
users may find the extra baggage of our Pointset, Generator, and 
Trajectory classes too much to just solve small ODE systems with minimal 
fuss. I guess I'd like to hear feedback about that stuff. We have lots of 
working demos in the download for people to try out....

My collaborator, Erik Sherwood, writes the PyDSTool SWIG interfaces and 
the C or Fortran wrappers to legacy ODE integrator codes. His ODE solver 
classes are themselves wrapped inside those of PyDSTool, and as such his 
python interface is more along the lines of the present SciPy ODE solvers 
(input and output is with Numarray arrays, not my bells-and-whistles 
Trajectory classes). Perhaps that would be the way to go in the short 
term. Certainly, he has done great work in wrapping high-end solvers like 
Hairer & Wanner's Dopri 853 and Radau 5 codes to work in Python, which 
could be very useful for SciPy users. He has even added support for 
user-defined events and non-constant mass-matrix support in Radau, among 
other useful extensions. He is in the process of adding more H&W 
integrators that use the same interface style. But you'd have to ask him 
if he'd like to be responsible for maintaining SciPy-released code!

Of course, the neat thing I think about PyDSTool wrapping these legacy 
integrators is that it doesn't require the user to write any actual _code_ 
in order to get super-fast ODE integration of C vector fields (Cf. 
VODE which uses python function callbacks for the right-hand sides). We 
use distutils to compile a DLL on the fly, which contains the v.f. and an 
instance of the integrator. This is automatically generated and reloaded 
by PyDSTool.

Last, but far from least, we have not yet migrated to new SciPy, so that 
would also be a sticking point. Our code uses a mixture of numarray and 
old SciPy array functions and classes, and needs some work to port that 
stuff to the new SciPy. Don't know when that will happen....

Sorry for the long post, but often the shortest questions require the 
longest answers!

Rob


On Tue, 7 Feb 2006, Alan G Isaac wrote:

> 1. Is PyDSTool destined to be part of SciPy? This seems
>   like a good idea ...
> 2. Was PyDSTool's scipy_ode.py forked from SciPy's ode.py
>   for any reason other than lack of communication?
>
> Cheers,
> Alan Isaac
>
> PS The note in scipy_ode.py reads::
>
>    # This is a trivially-adapted version of the original
>    SciPy integrate code.
>    # Changes: (1) removed the annoying print statement whenever
>    #       the ODE code is 'found', which is unwanted screen output during
>    #       computations.
>    #   (2) Made ode a new-style class.
>    #   (3) Changed a lambda occurrence in one of the methods to use noneFn instead.
>
>
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.net
> http://www.scipy.net/mailman/listinfo/scipy-user
>

-----------------------------------
Rob Clewley
Research Associate
Department of Mathematics
  and Center for Applied Mathematics
Cornell University
Ithaca, NY 14853
www.cam.cornell.edu/~rclewley
Tel: 607-255-7760
Fax: 607-255-9860