[SciPy-Dev] working on integrate.odeint

[Benny Malengier]

2013/8/23 Juan Luis Cano <juanlu001 at gmail.com>

>  On 08/23/2013 09:32 AM, Benny Malengier wrote:
>
>
>
>
> 2013/8/23 Juan Luis Cano <juanlu001 at gmail.com>
>
>> On 08/23/2013 08:42 AM, Juan Luis Cano wrote:
>> > On 08/23/2013 03:10 AM, Warren Weckesser wrote:
>> >> The other python file (test_lsoda.py) also needs the correction to the
>> >> coefficient.
>> >>
>> >> Warren
>> >
>> > Warren, thank you so much for pointing that out - I should never code
>> > at 2 AM. Now test_lsoda.py produces the correct output, which suggests
>> > I have to tune the relative and absolute error in test_lsoda2.py.
>>
>> Right, this was exactly the problem. Now the three tests produce the
>> same output and I understand much better the interface. In the branch I
>> pointed out before I wrote some issues as a to-do list and I plan
>> working on fixing some integrate.odeint problems in the following
>> months, when time and circumstances allow for that.
>>
>
>  It is nice to clean up odeint or lsoda, but those are really deprecated
> functions, replaced by sundials many years ago by Alan Hindmarsh (
> http://history.siam.org/oralhistories/hindmarsh.htm). As I mentioned
> before on this list, best would be to deprecate and replace by the modern
> version which still sees bug fixes (
> http://computation.llnl.gov/casc/sundials/main.html).
>
>  There are 3 python interfaces to sundials, one of which by me and a
> collaborator: https://github.com/bmcage/odes
>  The other 2 are readily found on the internet, and expose even more of
> the functionality.
>
>
> Well, I am farily new to contributing to SciPy so I missed those past
> conversations, but I did some googling and found that you created
> scikits.odes. I don't know if there were some conclusions about ode/odeint
> in SciPy, and though I find value in replacing old code with new one, I
> cannot tell if odepack is deprecated or not (deprecated for who?) and at
> least with these little steps I can fix the existing problems without
> disturbing much while trying to find a new ground-breaking solution. Unless
> a new discussion raises and it's decided that we can, say, incorporate
> scikits.odes into SciPy, or any other Python interface to some modern DE
> solvers. I don't have a strong opinion on this, I just know that odeint has
> some flaws right now and want to scratch my itch.
>

Well, I'm ok with that.

odepack, lsoda, vode, ... are deprecated in the sense that the authors
created the sundials package with more versatile and more tuned code. I'm
aware of yearly bug fixes to sundials, but not to the netlib stored old
fortran methods.

The old methods will off course work good in many cases. The point is, they
will not work ok in _some_ cases, that sundials will solve correctly. I
think the documentation should indicate the methods scipy uses are no
longer supported, and have been replaced by the sundials package of which
different python interfaces exist.

Sundials is used a lot in commercial and open source programs, just a short
list here: http://sundials.wikidot.com/projects





> Juan Luis
>
> _______________________________________________
> SciPy-Dev mailing list
> SciPy-Dev at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20130823/0f0465ab/attachment.html>