Mailman 3 February 2007 - SciPy-User

Proposal for more generic optimizers
by Matthieu Brucher Feb. 27, 2007

Feb. 27, 2007

Hi, I'm migratting toward Python for some weeks, but I do not find the tools I had to develop for my PhD in SciPy at the moment. I can't, for instance, find an elegant way to save the set of parameters used in an optimization for the standard algorithms. What is more, I think they can be more generic. What I did in C++, and I'd like your opinion about porting it in Python, was to define a standard optimizer with no iteration loop - iterate was a pure virtual method called by a optimize method -. This iteration loop was then defined for standard optimizer or damped optimizer. Each time, the parameters tested could be saved. Then, the step that had to be taken was an instance of a class that used a gradient step, a Newton step, ... and the same was used for the stoping criterion. The function was a class that defined value, gradient, hessian, ... if needed. For instance, a simplified instruction could have been : Optimizer* optimizer = StandardOptimizer</*some more parameters not relevant in Python*/(function, GradientStep(), SimpleCriterion(NbMaxIterations), step, saveParameters); optimizer->optimize(); optimizer->getOptimalParameters(); The "step" argument was a constant by which the computed step had to be multiplied, by default, it was 1. I know that this kind of writting is not as clear and lightweight as the current one, which is used by Matlab too. But perhaps giving more latitude to the user can be proposed with this system. If people want, I can try making a real Python example... Matthieu

2 2

Need Advice With Arrays and Calculating Eigenvectors
by Rich Shepard Feb. 27, 2007

Feb. 27, 2007

As a newcomer to python, NumPy, and SciPy I need to learn how to most efficiently manipulate data. Today's need is reducing a list of tuples to three smaller lists of tuples, creating a symmetrical matrix from each list, and calculating the Eigenvector of each of the three symmetrical matrices. The starting list contains 9 tuples. Each tuple has 30 items: a category name, a subcategory name, and 28 floats. These were selected from a database, and a tuple looks like this one: (u'soc', u'pro', 1.3196923076923075, 3.8109999999999999, 1.6943846153846154, 2.7393076923076922, 3.825538461538462, 5.0640769230769234, 3.609923076923077, 3.1429999999999998, 1.5936153846153849, 1.4893846153846153, 2.6563076923076929, 2.2156923076923074, 3.7973076923076921, 2.6884615384615387, 2.7008461538461543, 3.4992307692307687, 2.3813846153846154, 3.2199230769230769, 1.7726923076923078, 2.9855384615384613, 2.8829230769230771, 3.7862307692307695, 2.3791538461538462, 4.0949230769230773, 2.8703846153846153, 2.8296923076923073, 3.319230769230769, 1.8083076923076922) The three 'soc' subcategories need to have each of the 28 floats averaged and assigned to another tuple for 'soc'; same with the other two categories. That produces a list of three tuples, each with 29 items. Each of these 28 floats represents the average of a pair-wise comparison of values (in the non-numeric sense). So the first float above represents the cell (1,2), the second float represents the value of the cell (1,3) and so on. The diagonal of the matrix is 1. When I have these three symmetrical matrices, I want to call eigen() on each one to calculate the principal Eignevector. I can think of indirect ways of doing all this, but I'm sure that there are much more efficient approaches known to those who have done this before. So, I'd like your suggestions and recommendations. Of course, if I've not clearly explained my needs, please ask. Rich -- Richard B. Shepard, Ph.D. | The Environmental Permitting Applied Ecosystem Services, Inc. | Accelerator(TM) <http://www.appl-ecosys.com> Voice: 503-667-4517 Fax: 503-667-8863

2 6

Fitting - Gauss-normal distribution
by Marian Jakubik Feb. 26, 2007

Feb. 26, 2007

Hi, I am a SciPy newbie solving this problem: I would like to fit data with gaussian normal distribution.... First, I generated data: list=normal(0.00714,0.0005,140) Then I plot this data: pylab.hist(list,20) And at the end, I'd like to plot a gauss fit in the graph, also.... Could anyone help me, please? Marian This is my code: from numpy import * from RandomArray import * import pylab as p list=normal(0.00714,0.0005,140) p.hist(list,20) p.show()

6 5

ANN: new version of SFE (00.18.09) is available
by Robert Cimrman Feb. 26, 2007

Feb. 26, 2007

Hi, version 00.18.09 brings an improved problem definition syntax (regions are now per term, e.g. dw_poisson.Omega( ... )) and Navier-Stokes terms, see examples at http://ui505p06-mbs.ntc.zcu.cz/sfe cheers, r.

1 0

question re odeint
by Lev Givon Feb. 26, 2007

Feb. 26, 2007

While using scipy.integrate.odeint recently, I noticed that it modifies the object passed as the system's initial value. Is this behavior intentional? It seems that it would be preferable for odeint to use a copy of the initial value rather than manipulating whatever is passed as the second parameter directly. L.G.

2 1

Failed scipy 0.5.2 tests on IA64 Intel platform
by Ronald Jones Feb. 25, 2007

Feb. 25, 2007

Hello, I have recently installed numpy1.0.1/scipy0.5.2 on a IA64 machine using the intel compiler(v9.1.043) with the mkl (v8.1.014) libraries. The code compiles and runs. But six of the scipy.test(1) tests fail. Is this a known problem? There is also a problem with scipy.linsolve.umfpack. Is this significant? I tried the 19 Feb svn version of scipy and it failed 10 tests. The output of the scipy 0.5.2 tests are attached. regards, Ronald -- Ronald Jones iVEC, 'The hub of advanced computing in Western Australia' 26 Dick Perry Avenue, Technology Park Kensington WA 6151 Australia Phone: +61 8 6436 8633 Fax: +61 8 6436 8555 Email: ronald.jones(a)ivec.org WWW: http://www.ivec.org

1 0

how to uninstall Mac OS X ScipySuperpack Intel?
by Jason Addison Feb. 25, 2007

Feb. 25, 2007

I installed ScipySuperpack-Intel and have had some trouble. I thinking about just giving up on it and compiling my own. Before I do, I'd like to clean up from the packages that I installed: PyMC-1.1-py2.4-macosx10.4.mpkg gfortranCompleteInstaller.mpkg matplotlib-0.87.7-py2.4-macosx10.4.mpkg numpy-1.0.2.dev3522-py2.4-macosx10.4.mpkg scipy-0.5.3.dev2630-py2.4-macosx10.4.mpkg It looks like these install things deep into the filesystem, but I'm not sure what and where. Is there an easy way to uninstall? Is there a hard way? I didn't see any READMEs or such. Thanks ... jra

5 4

Re: [SciPy-user] Carateristic distance in a cloud of points
by Anand Patil Feb. 25, 2007

Feb. 25, 2007

>Correcting a stupid mistake. It bothered me to leave it, sorry for the >noise. > >Ga?l > >On Sun, Feb 25, 2007 at 01:03:03AM +0100, Gael Varoquaux wrote: > > >>Interesting remarks. You forced me to think a bit more about what I was >>trying to achieve. >> >> > > > >>What I am trying to do is to find out the right size to use for symbols >>when used on a 3D cloud of points. I am not sure what the right "typical >>distance" should be used. If those symbols are arrows then it seems that >>should be smaller than the typical inter-point distance. I have in mind >>something like this: >> >> > > > >>if you have n points, find out the distribution of distances, divide it >>by n**3, then take the value at 0.2 from the smallest. >> >> > > > >>I am having diffculties expressing my point, but the idea would be to >>consider that the typical distribution will increase as n**3 (which is >> >> > n**(1/3.) > > >>not obvious, for instance if the points are along a plane) and take the >>lower tail of the distribution, as we are interested in having symbols >>smaller than the inter-point distance. Taking not the smallest value, but >>the value at "20%" from the bottom helps getting rid of singular >>situations where a few points are very close but the major part is spread >>out. >> >> Hi Gael, Scaling relative to the cloud of points might be an easier way to go than scaling relative to the actual interpoint spacing. It would make sure your arrowheads are readable on the plot, even though some may appear oversized relative to their shafts. Also, points that are far apart in 3d space might appear close when viewed from particular angles, like 'optical doubles' in astronomy. If you want to make your symbols roughly the right size relative to the whole cloud, you might like a widely used quick-and-dirty method from statistics.: from scipy import * from numpy.linalg import eigh points = 2.*randn(100,3) C = cov(points.transpose()) D,V=eigh(points) Then the `error ellipse', the ellipsoid that kind of sort of tries to fit the point cloud, has major axes given by the columns of V with length equal to the sqrt of the corresponding elements of D. You could then calculate approximately how big or small the point cloud looks by projecting the major axes into your viewing plane. Hope that helps... that would be a milestone for me, my first time actually helping someone else on a Python mailing list. Cheers, Anand

1 0

Optimization
by Lorenzo Isella Feb. 24, 2007

Feb. 24, 2007

Dear All, I am going through the Scipy manual and I am trying to reproduce, as an exercise to learn Python and Scipy, some nonlinear least-square optimization which I was able to carry out using another language. I am collecting some questions which hopefully will help me understand Python a bit better: (1) What is the difference between from pylab import * and import pylab? (2) the 2nd question may be a non-problem: in running some of the examples in the tutorial by Olinfant, I could not use the xplt module, no matter whether trying import scipy.xplt, from scipy.xplt import * et similia. I bumped into http://lists.debian.org/debian-science/2007/01/msg00007.html which may be the answer (I am running Debian on my box as well and I installed Scipy from the Debian repository). Can I install xplt by itself? BTW, I also visited http://www.scipy.org/Cookbook/xplt but even using from scipy.sandbox import xplt does not help. (3) Is there a way to have arrays starting with index 1 rather than zero in Python? As you can guess, I do not have a strong C background. (4) This is the main question: I am trying to fit some experimental data to a log-normal curve. I would like to follow the same steps as in the tutorial, but something seems to be going wrong. I cut and paste the code I am using and attach a .csv file so that one can reproduce step by step my work. (5) Finally, if I load scipy and then write, e.g., z=10.3, how is z handled? Is it a floating point number? What if, for instance, I need to have a very large number of significant digits because they do matter for some computation I want to run? Can I have the equivalent of format long [Matlab statement], so that every non-integer number is by default treated with a certain precision? Here is the code: #! /usr/bin/env python from pylab import plot, show, ylim, yticks from scipy import * import pylab # now I want to try reading some .csv file data = pylab.load("120km-TPM.csv",delimiter=',') vecdim=shape(data) # now I introduce a vector with the dimensions of data, the file I read print "the dimensions of data are" print vecdim # now very careful! in Python the arrays start with index zero. diam=data[0:vecdim[0],0] # it means: slice the rows from the 1st one (0) to the last one ( # (vecdim[0]) for the first column (0) print "the dimensions of diam are" print shape(diam) #plot(diam,data[:,1]) #show() # uncomment them to plot a distribution # 1st problem: if uncomment the previous two lines, I get a warning and until I close # the window, the script does not progress. # now I try performing a least-square fitting from scipy.optimize import leastsq x=diam # just a list of diameters y_meas=data[0:vecdim[0],1] # measured data, for example the 2nd column of the .csv file def residuals(p, y, x): A1,mu1,myvar1 = p err = y-log(10.0)*A1/sqrt(2.0*pi)/log(myvar1)*exp(-((log(x/mu1))**2.0)/2.0/log(myvar1)/log(myvar1)) return err def peval(x, p): return log(10.0)*p[0]/sqrt(2.0*pi)/log(p[2])*exp(-((log(x/p[1]))**2.0)/2.0/log(p[2])/log(p[2])) p0 = [50000.0,90.0, 1.59] print array(p0) # now I try actually solving the problem print "ok up to here" plsq = leastsq(residuals, p0, args=(y_meas, x)) print "ok up to here2" print plsq[0] print array([A, k, theta]) print "So far so good" which produces this output on my box: $ ./read-and-plot.py the dimensions of data are (104, 10) the dimensions of diam are (104,) [ 5.00000000e+04 9.00000000e+01 1.59000000e+00] ok up to here TypeError: array cannot be safely cast to required type Traceback (most recent call last): File "./read-and-plot.py", line 51, in ? plsq = leastsq(residuals, p0, args=(y_meas, x)) File "/usr/lib/python2.4/site-packages/scipy/optimize/minpack.py", line 266, in leastsq retval = _minpack._lmdif(func,x0,args,full_output,ftol,xtol,gtol,maxfev,epsfcn,factor,diag) minpack.error: Result from function call is not a proper array of floats. Many thanks Lorenzo

3 4

fmin_cobyla help-passing constraints and bounds
by Siddhartha Jain Feb. 24, 2007

Feb. 24, 2007

I am trying to solve an optimization problem with more than 500 variables. Will fmin_cobyla be able to solve such a large problem in a reasonable amount of time? If so, then can anyone help me as to how to pass the variable bounds and constraints to fmin_cobyla? Thank you

1 0