Mailman 3 August 2008 - SciPy-User

advice on stochastic(?) optimisation
by bryan cole 29 Aug '08

29 Aug '08

Hi, I'll looking for a bit of guidance as to what sort of algorithm is most appropriate/efficient for finding the local maximum of a function (in 2 dimensions), where each function evaluation is 1) noisy and 2) expensive/slow to evaluate. I'd welcome any suggestions for where best to start investigating this (text books, references, web-sites or existing optimisation libraries). I've no background in this field at all. cheers, Bryan

5 6

Getting coordinates of a level (contour) curve
by Mico Filós 29 Aug '08

29 Aug '08

Hi, I have a fixed bivariate gaussian density function, and I want to compute the coordinates (x(l), y(l)) of a given level curve, where l is the parameter of the curve. I can easily plot the level x curve with matplotlib, using function 'contour', but I have no idea about how to get its coordinates (something like an (N,2) array specifying the coordinates of N points along the curve). With fsolve I can find one of such points, but its is not enough :) Thanks in advance for your help, Best Mico

7 15

in-place add for sparse.lil_eye
by Tony S Yu 29 Aug '08

29 Aug '08

I'm not sure if this is a closed bug or not, but in-place adding for sparse.lil_eye *sometimes* raises: AttributeError: 'numpy.ndarray' object has no attribute 'append' In the code below, the error doesn't occur unless the sparse matrix is multiplied by some number (`a` below) AND the added matrix (`A_offdiag` below) has entries off the main diagonal. Normal adding works fine. I noticed the 0.7.0 changes noted "numerous bug fixes" for the sparse module, so this problem may already be fixed (possibly with the closing of Ticket#680); I just wanted to bring it up in case it hadn't. Cheers, -Tony #~~~~~~~~~~~~~~~~~~~~~~~~~~~ import numpy as np import scipy.sparse as sparse N = 15 a = 2. A_offdiag = sparse.lil_diags([np.ones(N)], [1], (N, N)) A_eye = a * sparse.lil_eye((N, N)) A_eye += A_offdiag #~~~~~~~~~~~~~~~~~~~~~~~~~~~ Specs: --------- scipy 0.6.0 numpy 1.1.1 python 2.5.1 os.x 10.5.3

2 1

pyqwt or matplotlib
by bernardo martins rocha 29 Aug '08

29 Aug '08

Hi everybody, I'm starting to write a program with PyQt to visualize some traces from a system of ODEs. I would like to plot one graphic for each variable inside this PyQt application...and I'm wondering which one is the best for it: matplotlib or pyqwt? I will read some files using PyTables and then I'll plot everything and some of these files are very big, so I need something fast and good. I've been using matplotlib for some small programs and it's very nice, powerful and beautiful. I've already embedded some matplotlib plots inside a PyQt application. But I have the impression that pyqwt is faster than matplotlib. Is it true? Is there another library for plotting that would do the job? Thanks! Bernardo M. Rocha

7 9

Thoughts on GUI development
by Eric Bruning 28 Aug '08

28 Aug '08

Hi Gael, et al.: Regrettably, I couldn't attend scipy this year, but have been enjoying everyone's slides. I'm not entirely sure what the purpose of this message is, but one item in your slides was quite relevant to what I've been working on this week. In your lightning talk on the interactive shell, you wrote: "What do we gain with GUIs? -Pretty look and feel -Doesn't make you more productive/richer: no economic or academic incentive" At a gut level, I disagree that GUIs don't make you more productive. I imagine that you and others probably disagree, too. After all, we must find *some* non-superficial value in them to go to the effort of writing them! For certain datasets, I've found that I can't do the analysis I want without a good GUI for browsing and tagging data. Since this involves syncing up plots and animations across 4D, the code is non-trivial. So, the snag comes at the point of trying to justify all the time you spend writing a GUI app; you hope for future efficiency in analysis, but the benefits are pushed into the future right along with publishable results (==riches, such as it is in academia). And much of the efficiency comes by implementing the hardest features: not just plots, but draggable, zoomable, taggable, animated, linked plots. Perhaps the best way to conclude is by saying thanks to all those that are pushing forward on the graphical toolkits. The faster it is to make an app, the less conflict we'll all feel when justifying time spent crafting GUIs. -Eric

5 7

Re: [SciPy-user] pyqwt or matplotlib
by Pierre Raybaut 28 Aug '08

28 Aug '08

> > Message: 4 > Date: Tue, 26 Aug 2008 14:27:59 +0200 > From: bernardo martins rocha <bernardo.rocha(a)meduni-graz.at> > Subject: [SciPy-user] pyqwt or matplotlib > To: scipy-user(a)scipy.org > Message-ID: <48B3F6CF.2010600(a)meduni-graz.at> > Content-Type: text/plain; charset=ISO-8859-1; format=flowed > > Hi everybody, > > I'm starting to write a program with PyQt to visualize some traces from > a system of ODEs. I would like to plot one graphic for each variable > inside this PyQt application...and I'm wondering which one is the best > for it: matplotlib or pyqwt? I will read some files using PyTables and > then I'll plot everything and some of these files are very big, so I > need something fast and good. > > I've been using matplotlib for some small programs and it's very nice, > powerful and beautiful. I've already embedded some matplotlib plots > inside a PyQt application. But I have the impression that pyqwt is > faster than matplotlib. Is it true? Is there another library for > plotting that would do the job? > > Thanks! > Bernardo M. Rocha Hi, That is not an impression: PyQwt is much faster than matplotlib and is often used precisely to analyse huge data sets (here is an example: http://pyqwt.sourceforge.net/images/meq.pdf -- simple plotting, but very effective). On the other hand, as you may know, matplotlib has *a lot* more features, but if you don't need them... Pierre

6 5

looking for a negative binomial distribution
by Chris Fonnesbeck 28 Aug '08

28 Aug '08

I notice in the scipy dev wiki that the negative binomial random number generator is broken. In particular, it appears to round the first parameter, which is incorrect -- any real number is valid, not just integers. As a result, the method bombs out when passing it a parameter value of n<1, since it gets rounded to zero. This is a bit worrying for a common distribution like the NB, and it makes me wonder about other random number generators in numpy/scipy. Does anyone else know of a stable random number generating library for python?

3 6

Re: [SciPy-user] Curve fitting
by David Lonie 28 Aug '08

28 Aug '08

Thanks for the guidance -- I got a very good fit for an a*b^x+c model. Curve fitting seems a lot less intimidating now than it did 2 days ago :) Thanks again, Dave On Thu, Aug 28, 2008 at 10:22 AM, Warren Weckesser <warren.weckesser(a)gmail.com> wrote: > Hi David, > > On Thu, Aug 28, 2008 at 8:56 AM, David Lonie <loniedavid(a)gmail.com> wrote: >> >> Thanks for the fast reply! I plugged in my data to that script and >> added some output, but I get an invalid fit out of it. The fit I got >> from oocalc is approx >> >> y = 32 - 0.85^x, and fmin is returning >> y = 12 - 0.65^x, which for some reason returns an array of non-numbers >> when I try to generate plot data? > > If c[0]=12 and c[1]=-0.65, then your function is not 12-0.65**x, > but rather 12 + (-0.65)**x, which is not defined when x is not > an integer. As Ryan suggested, try a better starting guess. > You know the function must be decreasing, so c[1] must be > between 0 and 1. If I use c_initial_guess = [10, 0.5], I get > c_final = [12.20289742 0.65819867] > This is still a bad fit; I think this sum is not a good model for > your data. You'll get a better fit with a simple scaled > exponential: c[0]*(c[1]**x) > > > Cheers, > > Warren > > > _______________________________________________ > SciPy-user mailing list > SciPy-user(a)scipy.org > http://projects.scipy.org/mailman/listinfo/scipy-user > >

1 0

Curve fitting and LaTeX output
by David Lonie 28 Aug '08

28 Aug '08

I'm using the scipy package to analyze data from my research. I'm running into a couple problems I'd like some help with. 2 questions: 1) Curve fitting -- I have found linregress for linear functions, polyfit for polynomial fits, and a description of using lstsq to fit data that linearly depends on x. Is there a way to fit a curve of the form, for example, y = a + b^x? I don't mind a RTFM response to this, just please let me know which FM to R, because I can't find anything :) 2) LaTeX output -- I remember using R for stats before, but I'd like to avoid going back to it if possible. IIRC it had the ability to output latex tables, etc. that could be inserted into a document via \input{} (I think). Is there a python module that provides a similar output interface? Thanks in advance, Dave

6 11

possible to get INF in divide by zero ?
by Stef Mientki 28 Aug '08

28 Aug '08

hello, I wonder if it's possible to get the MatLab behavior : divide by zero gives infinite (INF) ? I even wonder if something like INF exists in python / scipy / numpy. Can someone clarify this ? thanks, Stef Mientki

1 0