matplotlib-users Mailing List for matplotlib

===== Original message from Vidar Gundersen | 2005年10月10日:
> there are one slide (page 18) in a presentation 'Python for
> Scientific Computing' by Eric Jones, which shows a comparison
> of vector multiplication speed between matlab and python.
i found another presentation using the same material:
http://www.iwce.nanohub.org/python.htm
http://www.iwce.nanohub.org/talks/python/python_talk1.pdf
see pages 58-59.

===== Original message from Travis Brady | 2005年10月10日:
> I've googled "python vs. matlab" but most results seem to address
> differences in the sort of code one has to write to achieve the same
> ends, I'm interested in speed.
there are one slide (page 18) in a presentation 'Python for
Scientific Computing' by Eric Jones, which shows a comparison
of vector multiplication speed between matlab and python.
i think i found it here: http://www.python9.org/p9-jones.ppt
unfortunately the above link is broken, and i've
been unable to find these somewhere else.

Travis Brady wrote:
>All,
>
>Lately folks in my office have started inquiring about the possibility
>of porting some Matlab code to Python. They're mostly concerned about
>cost and scriptability, but they're worried that the resulting Python
>code would be very slow. The current Matlab version of the prime
>porting target takes about 2 hours to run start to finish. 
>
>So I'm looking to either run some tests to prove that Python can keep up
>or take someone else's results from similar tests and evangelize with
>those.
>I've googled "python vs. matlab" but most results seem to address
>differences in the sort of code one has to write to achieve the same
>ends, I'm interested in speed.
>
>Anyone have any links or ammunition for the fight?
>
>thanks,
>Travis
> 
>
Travis: There was a thread on this on Numpy-discussion a while back. 
Here's an excerpt:
http://aspn.activestate.com/ASPN/Mail/Message/numpy-discussion/2396301
Bottom line is that if you link the atlas libs, Numeric or numarray is 
comparable to Matlab for linear algebra (matrix multiply and eigenvector 
computation).
-Jeff
-- 
Jeffrey S. Whitaker Phone : (303)497-6313
NOAA/OAR/CDC R/CDC1 FAX : (303)497-6449
325 Broadway Web : http://www.cdc.noaa.gov/~jsw
Boulder, CO, USA 80305-3328 Office: Skaggs Research Cntr 1D-124

Hello Travis,
I worked about 7 years with Matlab, but at the end I gave up and I switched
to Python. This is mainly due to the fact that with new Matlab releases
(starting from R14) it is almost impossible to build a distributable
executable file. You can compile an application into an exe, but you have to
distribute such big files that I gave up. Moreover, GUI capabilities of
Python (wxPython) are much much (much)^Inf superior to Matlab ones.
Regarding your question:
> They're mostly concerned about
> cost and scriptability, but they're worried that the resulting Python
> code would be very slow.
This much depends on how you use Matlab. If you use highly specialized
mathematical routines, image processing, optimization routines and
everything that involves large matrix computations (inversion,
decomposition, system solving etc...), I would be in doubt on which software
I would use.
Loops, conditionals and friends are much much faster in Python than in
Matlab, even using JIT acceleration.
Both software have a fast learning curve (in my opinion).
In case of large matrix computations, I would try to compare Matlab
capabilities to Python Numeric/scipy.
However, if you could specify a little bit more which use of Matlab you do,
I can probably give some more advice (even if I am not an expert with
Numeric/scipy, so my Python routines may have to be optimized by someone
else ;-) ).
HTH.
Andrea.
"Imagination Is The Only Weapon In The War Against Reality."
http://xoomer.virgilio.it/infinity77

>>>>> "Joost" == Joost van Evert <joo...@gm...> writes:
 Joost> What does this mean in matplotlib practice? Which
 Joost> partitions does one put plotting scripts, that contain
 Joost> latex formulae, to prevent this error?
In practice, matplotlib uses your home dir to cache certain things,
though this location can be overridden).
I just updated CVS to use shutil.move in the texmanager and elsewhere,
so this problem will disappear in the next release.
JDH

On Mon, 2005年10月10日 at 12:17 -0600, Fernando Perez wrote:
> This could be considered a matplolib bug, or a python one, depending on your 
> perspective. os.rename() doesn't work across partitions, which makes it 
> (IMHO) worse than useless. You have to either protect on all uses against this 
> exception, or use shutil.move(), which does work across partitions:
What does this mean in matplotlib practice? Which partitions does one
put plotting scripts, that contain latex formulae, to prevent this
error? 
Greets,
Joost

All,
Lately folks in my office have started inquiring about the possibility
of porting some Matlab code to Python. They're mostly concerned about
cost and scriptability, but they're worried that the resulting Python
code would be very slow. The current Matlab version of the prime
porting target takes about 2 hours to run start to finish.=20
So I'm looking to either run some tests to prove that Python can keep up
or take someone else's results from similar tests and evangelize with
those.
I've googled "python vs. matlab" but most results seem to address
differences in the sort of code one has to write to achieve the same
ends, I'm interested in speed.
Anyone have any links or ammunition for the fight?
thanks,
Travis
--=20
 Travis Brady
 td...@fa...
--=20
http://www.fastmail.fm - Same, same, but different=85

Joost van Evert wrote:
> Dear list,
> /usr/local/lib/python2.4/site-packages/matplotlib/texmanager.py
> in make_dvi(self, tex, force)
> 125 # dir and move it if necessary and then cleanup
> 126 if os.path.exists(dvibase):
> --> 127 os.rename(dvibase, dvifile)
> 128 for fname in glob.glob(prefix+'*'):
> 129 os.remove(fname)
> 
> OSError: [Errno 18] Invalid cross-device link
This could be considered a matplolib bug, or a python one, depending on your 
perspective. os.rename() doesn't work across partitions, which makes it 
(IMHO) worse than useless. You have to either protect on all uses against this 
exception, or use shutil.move(), which does work across partitions:
Definition: shutil.move(src, dst)
Docstring:
 Recursively move a file or directory to another location.
 If the destination is on our current filesystem, then simply use
 rename. Otherwise, copy src to the dst and then remove src.
 A lot more could be done here... A look at a mv.c shows a lot of
 the issues this implementation glosses over.
So if you think that Python is OK in having such a useless os.rename, then 
it's a matplotlib bug (this is, after all, known os.rename behavior). In the 
short term, though, matplotlib would be well advised to use only os.rename 
when doing renamings *known to be in the same directory*. Anytime you use 
rename to actually move things, this problem can appear (and it's 
configuration-specific, hence impossible to detect during in-house testing).
Cheers,
f

Alex,
It would be helpful if you post your basedir (from the beginning of=20
setupext.py),
the output of the build command (maybe just the lines for nc_image) and=20=
the operating
system you are using. This will show whether python is looking for=20
libpng in
the correct directory.
Nadia
On Oct 10, 2005, at 5:38 AM, Alexander Borghgraef wrote:
> On 10/6/05, John Hunter <jdh...@ac...> wrote:
>>
>> The place to start poking around is in setupext.py's build_image
>>
>> This function does a numerix switch to add flags for Numeric and
>> numarray if it finds them.=A0=A0Both paths (Numeric and numarray) =
call
>>
>> =A0=A0=A0=A0add_agg_flags(module)
>>
>> and add_agg_flags calls
>>
>> =A0=A0=A0=A0module.libraries.append('png')
>>
>> so libpng should be in your link chain.=A0=A0Since you have a =
nonstandard
>> Numeric install, make sure the numerix logic is working as it should
>> (eg what is the value of "numerix" in build_image?)
> =A0The value of numerix is=A0 "Numeric", as it should be.=A0 Also,=A0=20=
> add_agg_flags is called four times,
> of which once inside build_image, so libpng should be linked in. I=20
> tried linking libpng into a simple
> c++ program, no problem there.
>
> --=20
> Alex Borghgraef=

Dear list,
activating latex rendering by change of the matplotlibrc variable
'text.usetex' to true gives me the following error. I am using
matplotlib 0.84 using python 2.4.1 on linux. Does anyknow know how to
solve this? This problem doesn't appear for the simple plot command:
 from matplotlib.pylab import *
 title(r'$\rho$')
 plot([1,2,3,4])
 show()
But when I plot multiple lines using the following approach:
 from matplotlib import pylab
 def plotroc(x,y, label=None):
 pylab.figure(0)
 pylab.plot(x,y,label=label)
 pylab.xlabel('1-specificity')
 pylab.ylabel('sensitivity')
 pylab.axis(xmin=0, xmax=1, ymin=0, ymax=1)
 
 plotroc([1,2,3,4],[4,4,2,1],label='a')
 plotroc([1,2,3,4],[4,3,2,1],label='b')
 pylab.savefig('test.eps') 
 
matplotlib comes with the following message:
 exceptions.OSError Traceback
 (most recent call last)
 
 /usr/local/lib/python2.4/site-packages/matplotlib/backends/backend_gtk.py in expose_event(self, widget, event)
 316 x, y, w, h = self.allocation
 317 self._pixmap_prepare (w, h)
 --> 318 self._render_figure(self._pixmap, w, h)
 319 self._need_redraw = False
 320
 
 ...
 
 /usr/local/lib/python2.4/site-packages/matplotlib/texmanager.py
 in make_dvi(self, tex, force)
 125 # dir and move it if necessary and then cleanup
 126 if os.path.exists(dvibase):
 --> 127 os.rename(dvibase, dvifile)
 128 for fname in glob.glob(prefix+'*'):
 129 os.remove(fname)
 
 OSError: [Errno 18] Invalid cross-device link
 
Thanks in advance,
Joost

On 10/6/05, John Hunter <jdh...@ac...> wrote:
>
>
> The place to start poking around is in setupext.py's build_image
>
> This function does a numerix switch to add flags for Numeric and
> numarray if it finds them. Both paths (Numeric and numarray) call
>
> add_agg_flags(module)
>
> and add_agg_flags calls
>
> module.libraries.append('png')
>
> so libpng should be in your link chain. Since you have a nonstandard
> Numeric install, make sure the numerix logic is working as it should
> (eg what is the value of "numerix" in build_image?)
The value of numerix is "Numeric", as it should be. Also, add_agg_flags is
called four times,
of which once inside build_image, so libpng should be linked in. I tried
linking libpng into a simple
c++ program, no problem there.
--
Alex Borghgraef