Mailman 3 October 2004 - SciPy-User

Re: integrate.odeint
by Greg Novak 18 Oct '04

18 Oct '04

> Subject: [SciPy-user] integrate.odeint question > The documentation doesn't seem to give much information on how this > function works. Does anyone have pointers to what numerical methods > are implemented by this function and what the optional parameters mean > in detail? Speaking of integrate.odeint, I have a feature suggestion. Supplying a Python function to compute the derivatives of the ode system is great for quick, undemanding problems, but the Python function call overhead seems to completely, utterly kill performance for demanding applications. What I'd like to see is a function exactly like odeint (maybe called c_odeint), except instead of a Python function you give it a string with C code that defines a function that computes the derivatives. C_odeint will compile the function with Weave and then you can do the whole integration in C. This is much faster, in my experience. I have code to do this. I don't know if everyone will like the way I've set it up (it's basically the first thing I've written that involves Python), but I'm happy to send it along. Anyone who's interested in this, please let me know. Greg

4 3

multiplying matrices
by David Grant 16 Oct '04

16 Oct '04

Is there any way to multiple a whole bunch of matrices together easily? Ideally, the * operator would be for matrix multiplication as opposed to element-by-element multiplication. I've been doing this: listofmatricies = [blah, blah, blah, ....] return reduce(matrixmultiply,listofmatricies) Seems to work, but it would nice to be able to do it in one line as in matlab. -- David J. Grant <http://www.davidgrant.ca:81> Get Firefox! <http://spreadfirefox.com/community/?q=affiliates&id=1621&t=70>

3 3

Re: SciPy-user Digest, Vol 14, Issue 15
by Greg Novak 16 Oct '04

16 Oct '04

> From: Travis Oliphant <oliphant(a)ee.byu.edu> > Subject: Re: [SciPy-user] Re: integrate.odeint > > We have been interested in something like this for a while. I would > like to see the user be able to write c and/or fortran code to compute > the function to be integrated. I've collected the code that I wrote for this along with a brief explanation and put it here, in case anyone's interested. http://www.ucolick.org/~novak/integrate.tgz Greg

1 0

Re: [SciPy-user] multiple windows in gplt
by krivilli＠unberwoot.net 16 Oct '04

16 Oct '04

HI, -Maybe what u're looking for is gplt.hold('on') Chris Fonnesbeck wrote: > My Markov chain Monte Carlo code calls gplt.plot repeatedly to > generate histograms and traces of simulated values. However, rather > than re-use the sample gnuplot window, it generates a new frame for > each plot it produces, often leaving my screen cluttered with hundreds > of plots for more complicated models. (...) > Here is a sample of my plotting method, which uses > gplt.plot: > > def time_series(self,data,name,xlab='Time',ylab='Value',suffix='',same_axes=True): > 'Internal plotting specification for handling nested arrays' > > 'If there is only one data array, go ahead and plot it ... ' > if len(shape(data))==1 or same_axes: > > print 'Plotting',name > plot(data) > > 'Plot options' > grid('off') > xtitle(xlab) > ytitle(ylab) > > 'Save to file' > output("%s%s.png" % (name,suffix),'png') > #close() > > else: > '... otherwise plot recursively' > > tdata = swapaxes(data,0,1) > > for i in range(len(tdata)): > self.time_series(tdata[i],name+'_'+str(i),xlab,ylab,suffix) > regards, -- roberto (krivilli)

3 9

[novak@ucolick.org: Re: integrate.odeint]
by Greg Novak 16 Oct '04

16 Oct '04

I have an extremely unfocused question. I've written a bit of code that uses scipy to compute Lyapunov exponents for a particular dynamical system. The only subtlety (mentioned in a recent e-mail) is that I don't use integrate.odeint() but rather my own home-brew routine that functions like odeint, but instead of taking a python function it takes a string with C code that it compiles (along with an ode integrator from Numerical Recipes) with Weave so that the whole integration happens in C rather than calling back and forth between C/Fortran and Python. Here's the problem: while running last night, this program managed to fill 2 GB of memory and crash. There's no way that 2 GB of memory was required at any given time, so it must be leaking. This is my first real effort in Python, so facing this problem I feel a bit like a babe in the forest. Python has garbage collection and therefore memory leaks must be buried in the "inner workings" of the langauge. Therefore, I'm soliciting advice from the wise men and women of the scipy/python community: what's your strategy for tracking down bugs like this? What sorts of bugs are typical? If I were using the C/Python API directly, I would conclude that I'd messed up reference counts. But the reference counting increment/decrement macros don't appear when you use Weave. So, that's it. Any advice is welcome. Greg

3 2

Re: [SciPy-user] multiple windows in gplt
by krivilli＠unberwoot.net 15 Oct '04

15 Oct '04

HI, Chris Fonnesbeck wrote: > On Thu, 14 Oct 2004 20:40:18 -0500 (CDT), krivilli(a)unberwoot.net > <krivilli(a)unberwoot.net> wrote: > >>-This is different then use time_series() two times for two different data >>sets. Something like: >> >>class time_series(params,...): >> def __init__(): >> . >> . >> . >> >> def plot(data): >> #same code for ploting here >> . >> . >> . >> >>MyClass = time_series(foo,...) >> >>MyClass.plot(data1) >>MyClass.plot(data2) >> >>-This isn't quite the same. > > > Actually, I only have one instance of a "Plotter" class, that gets > passed around to objects that need it to plot for them. So it is more -Still, would be a different scope. Hence different than the example i gave which plots everthing on time_series() (no recursive). -Using recursive code, gplt will open a pipe for gnuplot (i think it's pipelined, not sure though) each time. The way it's now, is similar to open more than one gnuplot interactive session. -Using the method replot() (mentioned by Fernando Perez) would be a way to do what u want or use the other example i gave u. > like: > > class Plotter: > > def plot(self, data): > > class Thing: > > def plot(self, plotter): > > plotter.plot(self.time_series) > > p = Plotter() > > t1 = Thing() > t2 = Thing() > > t1.plot(p) > t2.plot(p) > regards, -- roberto (krivilli)

1 0

RE: [SciPy-user] Help needed: Problem in translating a file readroutine from Matlab to Python-Scipy
by Guasina Luca 15 Oct '04

15 Oct '04

Thank you to Travis and Stephen for the help ! Now all work correcly. What I am trying to do now I to write my processed data in VTK format (in fact VTR - non uniform rectangular grid with various point and cell data). I am trying to use pyvtk: do you think is possible to write COMPRESSED files? Visualisation of post-processed files will pick up files in VTK format; now I am using ParaView but I am thinking to move to MayaVi, directly from Python... Luca Guasina

2 1

integrate.odeint question
by matt 15 Oct '04

15 Oct '04

The documentation doesn't seem to give much information on how this function works. Does anyone have pointers to what numerical methods are implemented by this function and what the optional parameters mean in detail? -Matt B

2 1

scipy_distutils documentation
by tom fogal 15 Oct '04

15 Oct '04

Hi all, we've got some fortran code wrapped via f2py that we're probably be going to be adding to and using in a multitude of projects. I'd like to make a python package out of the sources, for ease of building / upgrading / use. Thus I'm trying to figure out how scipy_distutils works so I can use that to build the package. I can't seem to find any documentation on it, on the website or otherwise. Does anybody have a guide to setting up a project to use scipy_distutils perhaps? Or maybe there is another build system I should be looking into? Thanks, -tom

2 1

multiple windows in gplt
by Chris Fonnesbeck 14 Oct '04

14 Oct '04

My Markov chain Monte Carlo code calls gplt.plot repeatedly to generate histograms and traces of simulated values. However, rather than re-use the sample gnuplot window, it generates a new frame for each plot it produces, often leaving my screen cluttered with hundreds of plots for more complicated models. I have tried removing the gplt.close() statement in my plotting routine, but then the single window generated does not refresh with each new plot; the initial plot remains displayed until the end of the simulation. I know that if I call plot repeatedly from an interactive python session, subsequent plots are displayed properly in the same frame. Any ideas what may be causing this? Here is a sample of my plotting method, which uses gplt.plot: def time_series(self,data,name,xlab='Time',ylab='Value',suffix='',same_axes=True): 'Internal plotting specification for handling nested arrays' 'If there is only one data array, go ahead and plot it ... ' if len(shape(data))==1 or same_axes: print 'Plotting',name plot(data) 'Plot options' grid('off') xtitle(xlab) ytitle(ylab) 'Save to file' output("%s%s.png" % (name,suffix),'png') #close() else: '... otherwise plot recursively' tdata = swapaxes(data,0,1) for i in range(len(tdata)): self.time_series(tdata[i],name+'_'+str(i),xlab,ylab,suffix)

1 0