scientific libraries for python

[Harold Fellermann]

Hi all,

I want to use the current need for a Levenberg-Marquardt least squares
fitting procedure
for my long term desire to dive into scientific libraries for python.
However, I am always
confused by the shear sheer variety of available packages and the fact
that some of them
(Numeric, Numarray) seem to be outdated.

http://wiki.python.org/moin/NumericAndScientific gives a nice overview
of the most
popular packages. According to this listing, ScientificPytho n and
PyDSTool seem most
appropriate for what I generally work on (simulations of dynamical
systems and data
analysis).

Before I start to dive into one of these packages (I would go for
ScientificPytho n from what
I know so far), I want to ask about your experiences. Is there a good
reason to choose
one instead of the other? Or do they even work together, in the sense
that I can use
PyDSTool to generate data to be later analyzed by ScientifPython
without much timeconsuming
conversion a.s.o. in between?

Thanks for any suggestions!

- harold -

[Robert Kern]

Re: scientific libraries for python

Harold Fellermann wrote:[color=blue]
> Hi all,
>
> I want to use the current need for a Levenberg-Marquardt least squares
> fitting procedure
> for my long term desire to dive into scientific libraries for python.[/color]

Does it have to be Levenberg-Marquardt, or will another non-linear least squares
procedure do?
[color=blue]
> However, I am always
> confused by the shear sheer variety of available packages and the fact
> that some of them
> (Numeric, Numarray) seem to be outdated.[/color]

Fair enough.
[color=blue]
> http://wiki.python.org/moin/NumericAndScientific gives a nice overview
> of the most
> popular packages. According to this listing, ScientificPytho n and
> PyDSTool seem most
> appropriate for what I generally work on (simulations of dynamical
> systems and data
> analysis).[/color]

I find this page to be a little bit richer:

http://www.scipy.org/Topical_Software

[color=blue]
> Before I start to dive into one of these packages (I would go for
> ScientificPytho n from what
> I know so far), I want to ask about your experiences. Is there a good
> reason to choose
> one instead of the other?[/color]

They are largely orthogonal to one another in capabilities. You would use the
package that solves the problem in front of you. Quite possibly, you will be
using both.
[color=blue]
> Or do they even work together, in the sense
> that I can use
> PyDSTool to generate data to be later analyzed by ScientifPython
> without much timeconsuming
> conversion a.s.o. in between?[/color]

Both of them currently use the old Numeric so there are no real issues
converting between the two of them; they both use the same arrays. Install the
latest version of old Numeric (24.2), and you can convert to numpy arrays
without wasting memory.

For nonlinear least squares, I would probably recommend using my wrappers of
ODRPACK currently residing in the scipy sandbox. It uses numpy. It is very flexible.

http://projects.scipy.org/scipy/scipy/browser/trunk/Lib/sandbox/odr

If you need help installing scipy or using odr, come over to one of the scipy
lists, and I'll help you out.

http://www.scipy.org/Mailing_Lists

--
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
that is made terrible by our own mad attempt to interpret it as though it had
an underlying truth."
-- Umberto Eco