Mailman 3 March 2006 - SciPy-User

scipy.optimize.anneal with multiple parameters
by Tim Leslie March 5, 2006

March 5, 2006

Hi all, Before I dive too deeply into the internals of this code I thought I'd check here to see what people know. I'm trying to use simulated annealing to fit a quadratic to some data (I'll be using a more complex function later, the quadratic is just to get me started). I'm doing the following: func = lambda p: sum(abs(dat - array([p[0]*x*x + p[1]*x + p[2] for x in range(150, 450)]))) print anneal(func,[0.001, 0.001, 0.001],full_output=1,upper=[3.0, 3.0, 3.0 ],lower=[-3.0, -3.0, -3.0],feps=1e-6,maxiter=1000,schedule='fast') timl@mercury:~/thesis/src% python fit.py Traceback (most recent call last): File "fit.py", line 17, in ? print anneal(func,[0.001, 0.001, 0.001],full_output=1,upper=[3.0, 3.0, 3.0],lower=[-3.0, -3.0, -3.0],feps=1e-6,maxiter=1000,schedule='fast') File "/usr/lib/python2.4/site-packages/scipy/optimize/anneal.py", line 215, in anneal xnew = schedule.update_guess(x0) File "/usr/lib/python2.4/site-packages/scipy/optimize/anneal.py", line 92, in update_guess xc = y*(self.upper - self.lower) TypeError: unsupported operand type(s) for -: 'list' and 'list' I stole the syntax from the __main__ section of anneal.py and modified it to use a list of size 3, but I'm not sure if this is the correct way to do multiple parameters, the docs leave plenty to the imagination So, am I doing something wrong, in which case could someone show me the light, or is anneal a bit broken, in which case I'm happy to dive in and take a stab at fixing it. Cheers, Tim

2 3

scipy, blas, lapack, atlas and debian/ubuntu
by Cournapeau David March 5, 2006

March 5, 2006

Hi, I am using scipy for some time at work on my linux (x86 box), and it works fine; I am in general really happy about it as a matlab replacement. Recently, I tried to install the whole thing numpy/scipy/matplotlib on my minimac (again, linux, but on a ppc architecture), and I have some strange installation issues, which makes me wondering about the following issues on scipy installation. First, the conditions on both machines: ubuntu breezy, atlas3 (altivec or sse2 depending on the machine), numpy 0.9.5, scipy 0.4.6, matplotlib 0.87 ('stable' releases). - I am using ATLAS; to have a valid blas/lapack, I used the trick to rebuild a static LAPACK library using ATLAS. But I noticed this weekend that the liblapack.a included in the ubuntu ATLAS package is quite big (around 7 Mb), and the size of the packaged ATLAS liblapack.a and the one I am using are the same (by comparing the libraries, they have different md5, but when extracting them using ar x, the content looks like the same, ie same .o files). So I was wondering, does that mean the atlas lapack is actually a full implementation (maybe done by the debian/ubuntu packagers ?). Is there a way to be sure that a given library implements full lapack ? - When building numpy, is there a way to actually know which BLAS/LAPACK version is used ? For example, at work, on my x86 machine, python setup.py config gives me the following message at the end (I exported BLAS and LAPACK env variables to the location of the corresponding static versions, at /usr/lib/atlas/sse2) : """ lapack_info: Replacing _lib_names[0]=='lapack' with 'lapack' Replacing _lib_names[0]=='lapack' with 'lapack' FOUND: libraries = ['lapack'] library_dirs = ['/usr/lib/atlas/sse2'] language = f77 FOUND: libraries = ['f77blas', 'cblas', 'atlas', 'lapack'] library_dirs = ['/usr/lib/sse2', '/usr/lib/atlas/sse2'] define_macros = [('ATLAS_WITHOUT_LAPACK', None), ('ATLAS_INFO', '"\ \"3.6.0\\""')] language = f77 include_dirs = ['/usr/include'] """ I am not sure to understand the message: it seems like two lapack were found, which one is taken ? - Also, with BLAS/ LAPACK env variables, I select static libraries, but it seems like shared libraries are used afterwards: how can I know which is one is used ? I know how to do this kind of things with C programs (trakcing which libraries are loaded at runtime), but with python, it can quite hard to know which module uses which library. I hope I am not too unclear !, David

1 0

Table like array
by Michael Sorich March 5, 2006

March 5, 2006

Hi, I am looking for a table like array. Something like a 'data frame' object to those familiar with the statistical languages R and Splus. This is mainly to hold and manipulate 2D spreadsheet like data, which tends to be of relatively small size (compared to what many people seem to use numpy for), heterogenous, have column and row names, and often contains missing data. A RecArray seems potentially useful, as it allows different fields to have different data types and holds the name of the field. However it doesn't seem easy to manipulate the data. Or perhaps I am simply having difficulty finding documentation on there features. eg adding a new column/field (and to a lesser extent a new row/record) to the recarray Changing the field/column names make a new table by selecting a subset of fields/columns. (you can select a single field/column, but not multiple). merging tables (concatenate seems to allow a recarray to be added as new rows but not columns) It would also be nice for the table to be able to deal easily with masked data (I have not tried this with recarray yet) and perhaps also to be able to give the rows/records unique ids that could be used to select the rows/records (in addition to the row/record index), in the same way that the fieldnames can select the fields. Can anyone comment on this issue? Particularly whether code exists for this purpose, and if not ideas about how best to go about developing such a Table like array (this would need to be limited to python programing as my ability to program in c is very limited). Thanks, michael

5 7

Re: [SciPy-user] complex error function in scipy.special
by morovia March 3, 2006

March 3, 2006

Thanks for the replies. wofz(z)is the one I was looking for. On Fri, 03 Mar 2006 Arnd Baecker wrote : >On Fri, 3 Mar 2006, morovia wrote: > > > > > Hello, > > > > I would like to know whether complex error function which can accept complex argument is implemented in scipy.special. > > > > w(z) = i/pi*int(exp(-t**2)/(z-t))dt limits: -Inf to Inf > > > > where z is complex. > >What is implemented is this one: > >In [1]: import scipy.special >In [2]: scipy.special.erf? >Type: ufunc >String Form: <ufunc 'erf'> >Namespace: Interactive >Docstring: > y = erf(x) y=erf(z) returns the error function of complex argument >defined as > as 2/sqrt(pi)*integral(exp(-t**2),t=0..z) > >which is related to the above w(z), see >Abramowitz/Stegun, 7.1.3 and 7.1.4.: > http://www.math.sfu.ca/~cbm/aands/page_297.htm >However, I am not sure if using these relations is the best way. > >Wait - there is also > >In [6]: scipy.special.wofz? >Type: ufunc >String Form: <ufunc 'wofz'> >Namespace: Interactive >Docstring: > y = wofz(x) y=wofz(z) returns the value of the fadeeva function for >complex argument > z: exp(-z**2)*erfc(-i*z) > >So this looks like the one you are looking for. >(as usual: better compare the results for some known values). > >Best, Arnd > > >

1 0

complex error function in scipy.special
by morovia March 3, 2006

March 3, 2006

Hello, I would like to know whether complex error function which can accept complex argument is implemented in scipy.special. w(z) = i/pi*int(exp(-t**2)/(z-t))dt limits: -Inf to Inf where z is complex. Thanks, Morovia.

4 3

problem with odeint
by Ryan Krauss March 3, 2006

March 3, 2006

I am having a problem with odeint. My integration runs o.k. if I choose a fairly short time interval, but when I set the maximum value for my t vector higher than about 1 second, I get this message: In [178]: run sweptsine_ode.py lsoda-- at t (=r1), too much accuracy requested for precision of machine.. see tolsf (=r2) ls in above, r1 = 0.5448183650273E+00 r2 = NAN Excess accuracy requested (tolerances too small). Run with full_output = 1 to get quantitative information. Running with fulloutput produces: {'hu': array([ 0.00020682, 0.00043534, 0.00043534, ..., 0. , 0. , 0. ]), 'imxer': -1267232872, 'leniw': 23, 'lenrw': 68, 'message': 'Excess accuracy requested (tolerances too small).', 'mused': array([1, 1, 1, ..., 0, 0, 0]), 'nfe': array([19, 27, 31, ..., 0, 0, 0]), 'nje': array([0, 0, 0, ..., 0, 0, 0]), 'nqu': array([2, 3, 3, ..., 0, 0, 0]), 'nst': array([ 9, 13, 15, ..., 0, 0, 0]), 'tcur': array([ 0.00106956, 0.00235387, 0.00322454, ..., 0. , 0. , 0. ]), 'tolsf': array([ 7.88983883e-314, 7.88983883e-314, 7.88983883e-314, ..., 0.00000000e+000, 0.00000000e+000, 0.00000000e+000]), 'tsw': array([ 0., 0., 0., ..., 0., 0., 0.])} The script is attached and should be self contained. Messing with atol and rtol doesn't seem to help, so I don't know what the message about requested accuracy means. Any help/direction in fixing this would be appreciated. Thanks, Ryan

2 1

fftw and scipy.show_config() on SUSE 10.0
by Nils Wagner March 3, 2006

March 3, 2006

Hi all, I have installed the rpm's w.r.t. to fft's. /usr/lib64/libdfftw.so.2 /usr/lib64/libdfftw.so.2.0.7 /usr/lib64/libdrfftw.so.2 /usr/lib64/libdrfftw.so.2.0.7 /usr/lib64/libsfftw.so.2 /usr/lib64/libsfftw.so.2.0.7 /usr/lib64/libsrfftw.so.2 /usr/lib64/libsrfftw.so.2.0.7 /usr/lib64/libfftw3.so.3 /usr/lib64/libfftw3.so.3.0.1 /usr/lib64/libfftw3f.so.3 /usr/lib64/libfftw3f.so.3.0.1 but scipy.show_config() cannot find them. How can I fix this problem ? Do I need symbolic links ? Is /usr/lib64 not standard ? Nils Numpy version 0.9.6.2193 Scipy version 0.4.7.1630 dfftw_info: NOT AVAILABLE fft_opt_info: NOT AVAILABLE fftw2_info: NOT AVAILABLE fftw3_info: NOT AVAILABLE

1 0

Re: [SciPy-user] gfortran, ifc, compat-g77
by Nils Wagner March 3, 2006

March 3, 2006

Eric Emsellem wrote: > Hi, > I have compiled these on two machines: a laptop (see output of cpuinfo > below) > and a desktop both with Suse 10. With the procedure I sent you, I had > no problem installing all of it, except for the line to be added in > the setup of matplotlib (I hope the new version has been updated). > > And no, I cannot reproduce the failure you have with the test on > scipy. sorry... (I am not an expert there, so cannot really help) > > Eric > > processor : 0 > vendor_id : GenuineIntel > cpu family : 6 > model : 9 > model name : Intel(R) Pentium(R) M processor 1700MHz > stepping : 5 > cpu MHz : 598.146 > cache size : 1024 KB > fdiv_bug : no > hlt_bug : no > f00f_bug : no > coma_bug : no > fpu : yes > fpu_exception : yes > cpuid level : 2 > wp : yes > flags : fpu vme de pse tsc msr mce cx8 sep mtrr pge mca cmov > pat clflush dts acpi mmx fxsr sse sse2 tm pbe est tm2 > bogomips : 1197.31 > > >> >> Hi Eric, >> >> >> Finally I was able to compile ATLAS on SuSE 10.0. >> BTW, what is the output of cat /proc/cpuinfo ? >> >> numpy.test(1,10) works fine. >> Now I have some trouble with scipy.test(1,10). >> Can you reproduce this failure ? >> >> This is the output of scipy.test(1,10) >> >> snip >> >> check_y_stride (scipy.linalg.tests.test_fblas.test_zswap) ... ok >> check_algebraic_log_weight >> (scipy.integrate.tests.test_quadpack.test_quad) ... ok >> check_cauchypv_weight (scipy.integrate.tests.test_quadpack.test_quad) >> ... ok >> check_cosine_weighted_infinite >> (scipy.integrate.tests.test_quadpack.test_quad)STOP 778 >> >> Program exited with code 012. >> (gdb) bt >> No stack. >> >> >> Nils >> > Hi Eric, Did you use gfortran to compile lapack ? Robert mentioned that gfortran is unstable ! How did you compile ATLAS in detail ? Which version did you use 3.7.11 developer or 3.6 stable ? Which flags did you use -fPIC etc ? Also which fortran compiler did you use gfortran / g77 ? It would be very kind of you if you could expand on that. Thanks in advance. I look forward to hearing from you soon. BTW, did you try my code sparse_test.py ? Nils

1 0

minimizers don't work - d1mach problem
by Christian Kristukat March 2, 2006

March 2, 2006

Hi, sorry for posting this again, but I'd really like to have the minimizers work again. Every call to any of the miminizers scipy.fmin* fails with: Adjust D1MACH by uncommenting data statements appropriate for your machine. STOP 779 I already looked at Lib/special/d1mach.f but I don't see what I should change there. Anyway, d1mach.f hasn't been changed for ages so I doubt that the problem is there. Here's my configuration: P4 Python 2.4.1 (#1, Sep 13 2005, 00:39:20) [GCC 4.0.2 20050901 (prerelease) (SUSE Linux)] on linux2 numpy/scipy from svn ATLAS built with gfortran Regards, Christian

2 7

Matplotlib wiki cookbook link
by Ed Schofield March 2, 2006

March 2, 2006

Hi Andrew, It seems the matplotlib site has a link to http://www.scipy.org/wikis/topical_software/MatplotlibCookbook, which doesn't exist. Could you please configure the wiki so this URL that refers to the correct page (http://www.scipy.org/Cookbook/Matplotlib)? Or is it possible for normal wiki users to do this? -- Ed

3 3