Em Seg, 2008-12-08 às 12:25 -0200, Ronan Paixão escreveu:
> Em Seg, 2008-12-08 às 05:50 -0800, mabshoff escreveu:
> >
> > > > As I saw matplotlib/pylab being mentioned, in combination with
> > slowdowns, I
> > > > thought I'd mention that plotting in pylab itself goes
> > progressively slower.
> > > > On a system here, running the following:
> > >
> > > > import pylab
> > > > import time
> > > > n=300
> > > > m=100
> > > > for i in range(n):
> > > > start=time.time()
> > > > pylab.plot(range(m), [pylab.sin(pylab.pi*j/(m+0.0)) for j in
> > range(m)])
> > > > pylab.savefig("name%04i.eps"%i)
> > > > print "plotting took %f sec"%(time.time()-start)
> > >
> > > > Will print "plotting took 0.051967 sec" for the second run (first
> > goes
> > > > slower), but then the time to plot the next plots goes slowly up,
> > to about
> > > > 0.55 sec after 300 plots.
> > >
> > > Thanks. That is more than a 10-fold increase!
> > >
> > > Michael: could this contribute significantly to the slowness of the
> > > plot Fourier
> > > series you mentioned in the other thread?
> >
> I did some tests here, putting an os.system('free -m') in the loop and
> noticed that the memory usage keeps going up, which may explain the
> slowdown.
>
> However, more importantly, I noticed some other things in the tests. I
> Ctrl+C'ed the test and quit ipython, but my CPU was still at 100% usage.
> Investigating, I found that there were a bunch of gs processes running,
> converting files to .eps. Those files were the ones generated at the
> loop with savefig(). So, the problem really lies in that saving the plot
> is very slow and doesn't lock the interpreter (which is good), but it
> doesn't reveal where the bottleneck is. Then, doing a loop and putting a
> lot of those in the background will inevitably degrade performance
> either because of mem usage or because of increased CPU usage by
> background processes.
I just noticed one more thing which will also degrade performance of the
savefig() and increase memory usage. The code above never closes the
figure. Because of that, each plot is done above the previous plot
(which explains the differences of colors from plot to plot). That means
matplotlib is storing a curve for each time it plots, so in the last
plot, it's saving not one curve, but 300 of them, which can be seen as
an increase of the .eps file size.
Testing it, the problem is fixed by inserting a pylab.close() inside the
loop. The plot keeps the same time.
Ronan
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---
