[SciPy-user] array vs matrix, converting code from matlab

Bill Baxter wbaxter at gmail.com
Fri Apr 21 01:33:33 CDT 2006


One thing...  I'm not sure why you think porting Netlab to SciPy would be
such a huge task.  It's a big task, sure.  Porting to C++ would definitely be
a huge task.  But I would think porting to another high-level language like
python would be a one-summer job for a reasonably clueful grad student.
It's only 9000 lines of code excluding comments and blank lines:

[BAXTER-PC<7>netlab]> egrep -v '(^%|^$)' *.m | wc
   8971   41422  361276

I think converting 100 lines a day for 90 days is not unreasonable.  That
includes all the demos too.  If you leave out the demos its about half that:

[BAXTER-PC<18>netlab]> egrep -v '(^%|^\s*$)' `ls *.m | grep -v '^dem'` | wc
   4171   18725  156628

Ok maybe it's still a little unreasonable.  Alright, maybe it's not a 1-man
summer job.  I've also ignored testing and converting the comments, but the
task is also fairly parallelizeable.  Probably a little team of 3 eager new
grad students could do a bang up job over a summer.

--bb

On 4/21/06, David Cournapeau <david at ar.media.kyoto-u.ac.jp> wrote:
>
> > Bill Baxter wrote:
> > > By the way, I'd be interested in an n-dimension Gaussian function for
> > > NumPy/SciPy too.
> > >
> > > Anyone else interested in machine learning and or bayesian methods?  A
> > > port of Netlab ( http://www.ncrg.aston.ac.uk/netlab/index.php) in
> > > SciPy would be great.
> > Actually, I am porting a code for Gaussian Mixture Models with batch and
> > online EM. I first try to do a pure python version to get an idea on
> > scipy capabilities, and then I intend to create the stub to a C
> > implementation (which already exists for matlab, the core being
> > independant of matlab). I am  hoping to have a much cleaner
> > implementation, and more extensible (ie using other pdf, and why not
> > more general models) using python
> > languages capabilities (module, inheritance, etc...).
> >
> > I think porting netlab would be a huge task, and quite difficult; there
> > is also torch ( http://www.torch.ch/) which may be interesting to use
> > (C++ code, BSD license). Having a machine learning tool box would be a
> > step forward for scipy, I guess.
> >
> > David
> >
> > ____________
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.scipy.net/pipermail/scipy-user/attachments/20060421/a10da569/attachment.htm


More information about the SciPy-user mailing list