Numerical Programmers

[nardinc]

I am curious if, and how many numerical programmers there are in here?

Does anybody know of a good forum for discusing numerical related issues, like setting up LAPACK?

I don't know how these topics would be treated here since they don't realy pertain to Fedora.

Thanks.

[icy-flame]

If you are not restricted to LAPACK, then SciPy's mailing list is a good place to look at.

[mwette]

I do some of this stuff. I have installed LAPACK (yum install lapack lapack-devel).
The yum repo's have a lot. octave, R, python-numeric, ...

[nardinc]

Originally Posted by icy-flame

If you are not restricted to LAPACK, then SciPy's mailing list is a good place to look at.

Thank you, for the resource. I have never used Python, but SciPy looks quite interesting. Per haps I will check it out more later. Right now I have Matlab, that seems to be quite similar functionality.

[nardinc]

Originally Posted by mwette

I do some of this stuff. I have installed LAPACK (yum install lapack lapack-devel).
The yum repo's have a lot. octave, R, python-numeric, ...

If you have you used LAPACK, what did you use it for? Do you know of any online communities to discuss numerical issues?

I am interested in discussing things beyond LAPACK, but don't know if this is the oppropriate place.

There are many things that I would like to discuss.
* How fast does C++ code run compared to f77, f90
* Discus different "element free" methods (i.e. particle methods) for solving PDEs.

This is all that I can think of at the moment.

[icy-flame]

I dont specifically use LAPACK, and I dont have much interest in the computer science side of it. I use SciPy for engineering programming. So, I am using it, rather than developing it. But there are very active discussion and development regarding to numpy, of which SciPy based on. I use ATLAS library and FFTW as the foundation of my numpy build.

I like Python because it allows very rapid development cycle, and numpy is only a thin layer of wrapper on top of some of the most efficient libraries written in c/c++/fortran. SciPy is design to me MatLab equivalent.

My impression of this forum is more generic FC usage and support, perhaps not much into the core of the numerical algorithms. So I suggest you have a look at numpy.

[nardinc]

Originally Posted by icy-flame

I like Python because it allows very rapid development cycle, and numpy is only a thin layer of wrapper on top of some of the most efficient libraries written in c/c++/fortran. SciPy is design to me MatLab equivalent.

My impression of this forum is more generic FC usage and support, perhaps not much into the core of the numerical algorithms. So I suggest you have a look at numpy.

I am installing SciPy right now. You say that you use it for engineering programming, that is my intent also. What kinds of projects do you do?

[nardinc]

Do you use eclipse at all for you developement? I use it for Fortran/Java/C++, but I have never tried any of the Python pluggins.

[mwette]

Originally Posted by nardinc

If you have you used LAPACK, what did you use it for? Do you know of any online communities to discuss numerical issues?

I am interested in discussing things beyond LAPACK, but don't know if this is the oppropriate place.

There are many things that I would like to discuss.
* How fast does C++ code run compared to f77, f90
* Discus different "element free" methods (i.e. particle methods) for solving PDEs.

This is all that I can think of at the moment.

Not much lately. I have used QR routines in the past. By the way I just checked
octave. It is using "atlas" package, which is the hardware-tuned version of blas
lapack.

I did use python Numeric to bind in some custom LP solvers.

I used to follow the numerics a lot more. If you have access to USENET
(i.e., "news") you could see if the "sci.math.numerical" (don't remember
this quite right) group is still active. I have earthlink and they provide
access to some news servers. -- Matt

[EDIT: the news group is "sci.math.num-analysis"]

[icy-flame]

Originally Posted by nardinc

You say that you use it for engineering programming, that is my intent also. What kinds of projects do you do?

This thread might interests you, from where I made the following post:

Originally Posted by icy-flame

Condense matter physics, Acoustics, Laser generation and detection of Ultrasonic signals.

Electronics circuit simulation:
Qucs

Programs developed are mostly python based. (nedit is a perfect for a simple editor)
numpy
scipy
Gnuplot + Gnuplot.py
Veusz

Good drawing prog:
Skencil
Qcad
Xfig

Yes, I am coding something like datafusion, signal processing etc.

Originally Posted by mwette

I did use python Numeric to bind in some custom LP solvers.

FYI, Numeric is being replace by numpy, so if you are writting new codes, try to stick with numpy.

[nardinc]

I have downloaded all the goodies, numpy, scipy, ipython, and I am not sure exactly how to put them all together to do something simple; like plotting y=exp(x). Another thing to consider is that I have no experience with Python. Could one of you give me "The big picture" of how all of this fits together?

I am famillar with Matlab, how does the process compare with that?

I have looked online, and I am not sure where to start.

Thanks.

[mwette]

Hmmm. Good one. There should be examples in the source documentation.
You could check if there are any examples/docs in /usr/share/doc/<pkg-name>.

Also, I hate to give this to you, but another python package that you may want,
given that you know MATLAB is "matplotlib". This may be in the archives. It is
a python plotting package that is designed to be very much like MATLAB plotting.

Try the sourceforge site tutorial

[EDIT: yum package is "python-matplotlib"]

[icy-flame]

Originally Posted by nardinc

I have downloaded all the goodies, numpy, scipy, ipython, and I am not sure exactly how to put them all together to do something simple; like plotting y=exp(x). Another thing to consider is that I have no experience with Python. Could one of you give me "The big picture" of how all of this fits together?

I am famillar with Matlab, how does the process compare with that?

I have looked online, and I am not sure where to start.

Thanks.

If you like to have a taste of Python, there is a tutorial comin with Python's own doc. Though it is a bit too basic, but will get you started.

As of SciPy numpy and matplotlib, SciPy;s website contain quite a bit of info on those and beyond:
http://www.scipy.org/Cookbook
http://www.scipy.org/Documentation

Check the cookbooks first for a quick start. Then one or two tutorials of your liking under the Doc section for SciPy. Finally the numpy, or even the old Numeric tutorial is a good read for more advance and indepth numerical operations, since Numeric and numpy methods are largely compatible. Once you get to know what its all about, then the API doc is all you need.

[bryancole]

Python is a really nice platform for scientific/engineering computing. For plotting, you'll want the python-matplotlib (see http://matplotlib.sourceforge.net/). This provides a matlab-like set of plotting functions. For interactive work, IPython has a special "-pylab" mode (command line option), which automatically imports matplotlib and NumPy/Numeric. With this, you just start it and you can start plotting directly e.g.

>>> x=arange(0,10,0.1)
>>> y=cos(x)
>>> plot(x,y)

... and a plot window, with interactive zooming/panning appears.

Matplotlib provides great rendering quality (better than matlab IMHO) and you can export publication-quality firgures in a wide variety of formats (.png .jpeg .svg. .ps .eps). You can also embed matplotlib figures in your own GUI applications.

As far as IDEs go, I think Eclipse + pydev is the best FOSS option (although everyone has their own preferences here). Note, you may need Sun java installed to run pydev reliably.

Matplotlib doesn't yet provide much 3D support (there is a mpl3d module, but I've not tried it). For 3D visualisation, you can use VTK (www.vtk.org) with Python. Unfortunately, this isn't yet included in any repository (AFAIK) so you'll need to build this yourself. It's not difficult (but a bit boring, as VTK is very big). There are various higher level packages based on VTK/Python, such as Mayavi.

BC

[nardinc]

Thank you for the overview. That is exactly what I was looking for! I tried it and worked great. I imagine that after learning a little python I will be able to write little scripts and functions similar to m-files in Matlab?

With eclipse and pydev, do you have a similar interactive environment as with ipython -pylab?