Mailman 3 December 2005 - SciPy-User

Re: [SciPy-user] fmin_bfgs
by LOPEZ GARCIA DE LOMANA, ADRIAN Dec. 24, 2005

Dec. 24, 2005

Thanks for your answer. I've found the variable "rhok" at the optimize.py, and I would like to run it again setting a larger number, but, what is large? rhok = 1 / Num.dot(yk,sk) ------> rhok = 0.1?, rhok = 1000000.0? Thanks, Adrián. -----Original Message----- From: scipy-user-bounces(a)scipy.net on behalf of Travis Oliphant Sent: Wed 21/12/2005 06:53 To: SciPy Users List Subject: Re: [SciPy-user] fmin_bfgs LOPEZ GARCIA DE LOMANA, ADRIAN wrote: >Hi all, > >I'm testing the routine fmin_bfgs from scipy.optimize and I have some problems. I do not understand why it gives me this warning. > >Warning: Desired error not necessarily achieved due to precision loss <---------------------------- > Current function value: 152.114542 > Iterations: 5 > Function evaluations: 23 > Gradient evaluations: 13 > > > > This is a potential problem with the Quasi-Newton minimizers. If you look in the code at where this error shows up you will see that it happens because of a divide-by-zero problem. In practice either the gradient is getting updated or the function value isn't getting updated on an iteration. I'm not sure what to do in this situation as it theoretically shouldn't happen and so is perhaps due to the fact that the function is not able to change in sufficiently small ways (thus the precision-loss error). One approach is to reset the Hessian calculation and move on (but I did not like that result on some problems I was working with). Another approach is to set rhok = some large number and move on. You could edit optimize.py to do that. Right now, the code is not guessing what to do but stopping and letting you know of the failure of the quasi-Newton method on your function. I don't think anything is wrong with fmin_bfgs per-say (except for a better way to deal with this situation). >Does anyone face the same problem? Is there any problem with the fmin_bfgs routine? Why is stopping while the gradient norm is bigger than gtol? What does it mean that the precission was lost? > > The stoppage is occuring not because you've found a minimum but because the method has essentially failed. You could try again from a different starting location or try a different optimizer. There is no "always best" optimizer that I know of. Thus, the family of optimizers that exists in scipy. Try them and see if they work for you. Best, -Travis _______________________________________________ SciPy-user mailing list SciPy-user(a)scipy.net http://www.scipy.net/mailman/listinfo/scipy-user

4 3

polynomial division problems
by Ted Horst Dec. 24, 2005

Dec. 24, 2005

Polynomial division seems not to be working for me: >>> import scipy >>> scipy.__core_version__ '0.8.6.1693' >>> x2m1 = scipy.poly1d([1,0,-1]) >>> xp1 = scipy.poly1d([1,1]) >>> xm1 = scipy.poly1d([1,-1]) >>> x2m1/xp1 hangs, it seems the args in polydiv are backwards, so: >>> xp1/x2m1 [poly1d([ 0.]), poly1d([1, 1])] but even with args reversed, this should be poly1d([1, -1]) Hopefully this is an easy fix for somebody. Ted

3 4

Possible bug in loadmat and savemat use of fread and fwrite
by Souheil Inati Dec. 23, 2005

Dec. 23, 2005

Hi all, I am new to the scipy (and python). I have just gotten scipy working on a mac G4 OS X 10.4.4 with python 2.4.2, thanks to Chris Fonnesbeck's instructions. Apologies in advance for the long post and if I am asking a question previously answered on the mailing list. I can't figure out how to search the mailing list properly. Mailman itself (scipy.net) doesn't have a search and the plone interface to mailman on scipy.org does nothing for searching. Google is not terrible. Anyway.... I'm having problems with io functions loadmat and savemat. I've dug as far as my limited python knowledge can get me and my current working hypothesis is that there is a bug in the low-level fread and fwrite functions, or maybe the way in which the loadmat and savemat functions call them. The code in io/examples works fine and the reading and writing of binary files via a combination of fopen and fwrite works fine as well. Any help or nudge in the proper direction would be greatly appreciated. Thanks, Souheil --------------------------------- Souheil Inati, PhD Assistant Professor Center for Neural Science and Department of Psychology New York University 4 Washington Place, Room 809 New York, N.Y., 10003-6621 Office: (212) 998-3741 Email: souheil.inati(a)nyu.edu ------------------------------------ ------------------------------------ Here's the behavior that i see for loadmat I created a file in Matlab 7.1 >> A = randn(4) >> save -V6 foo A In python: from scipy import * b = io.loadmat('foo.mat') gives a bus error. Reading through io/mio.py at around line 744 in the the beginning of the loadmat function is this bit of code: fid = fopen(full_name,'rb') test_vals = fid.fread(4,'byte') If I try to run this directly from python, like this: from scipy import * fid = fopen('foo.mat','rb') test_vals = fid.fread(4,'byte') I get the same bus error. Using the lower level python file functions directly works: fid = file('foo.mat','rb') b = fid.fread(4) returns a string b = 'MATL' as expected. So presumable there is an error in fread. ------------------------------------ ------------------------------------ For the savemat function the problem looks like this: from scipy import * io.savemat('foo2.mat',{'a': array([1.,2.,3.])}) ------------------------------------------------------------------------ --- exceptions.TypeError Traceback (most recent call last) /Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site- packages/scipy/io/mio.py in savemat(filename, dict) 871 imagf = var.dtypechar in ['F', 'D'] 872 fid.fwrite([var.shape[1], var.shape[0], imagf, len (variable)+1],'int') --> 873 fid.fwrite(variable+'\x00','char') 874 if imagf: 875 fid.fwrite(var.real) /Library/Frameworks/Python.framework/Versions/2.4/lib/python2.4/site- packages/scipy/io/mio.py in write(self, data, mtype, bs) 219 howmany,mtype = getsize_type(mtype) 220 count = product(data.shape) --> 221 numpyio.fwrite(self,count,data,mtype,bs) 222 return 223 TypeError: argument 4 must be char, not None

2 1

boxcar problem
by John Hunter Dec. 22, 2005

Dec. 22, 2005

Any ideas on this one? In [1]: import scipy In [2]: scipy.__version__ Out[2]: '0.3.0_266.4239' In [3]: scipy.stats.boxcox(scipy.rand(100)) --------------------------------------------------------------------------- exceptions.ValueError Traceback (most recent call last) /home/titan/johnh/<console> /opt/lang/python-2.3.4/lib/python2.3/site-packages/scipy/stats/morestats.py in boxcox(x, lmbda, alpha) 279 return -boxcox_llf(lmb,data) 280 lmax = optimize.brent(tempfunc, brack=(-2.0,2.0),args=(x,)) --> 281 y, lmax = boxcox(x, lmax) 282 if alpha is None: 283 return y, lmax ValueError: too many values to unpack In [4]:

2 1

Re: [SciPy-user] Mac installers for scipy, etc.
by Brian Granger Dec. 22, 2005

Dec. 22, 2005

What dependencies are there for these (like fortran and fftw)? Also, how are you using gcc 4.0? I thought nobody had gotten it to work on the Mac yet? Thanks Brian Granger Brian Granger, Ph.D. Assistant Professor of Physics Santa Clara University bgranger(a)scu.edu Phone: 408-551-1891 Fax: 408-554-6965 >>> fonnesbeck(a)gmail.com 12/20/05 7:03 PM >>> For interested OS X users, I have provided mpkg installers for scipy_core and scipy from svn, as well as matplotlib 0.85.1 (cvs) and numarray 1.5.1. These were all built on OS X 10.4 on Python 2.4.2. http://trichech.us/mac Enjoy. -- Chris Fonnesbeck Atlanta, GA _______________________________________________ SciPy-user mailing list SciPy-user(a)scipy.net http://www.scipy.net/mailman/listinfo/scipy-user This message scanned for viruses and SPAM by GWGuardian at SCU (MGW1)

4 3

(almost all) lapack functions were already under 'scipy.linalg.flapack'
by Evan Monroig Dec. 22, 2005

Dec. 22, 2005

Hi, I am not sure if here is the right place, but here is a piece of experience that others might find useful. I started to port my simulation framework from gnu octave to python/scipy, because I thought performance would be greatly improved, without too much coding effort (I already know python). For this I needed to factorize a reasonably big (1800 * 200) matrix using the QR decomposition. So I tried the function 'scipy.linalg.qr', and found that it took *minutes* to compute it where octave could do that within a few seconds. So I looked on the internet, on the scipy website, google, etc. for one or two hours looking for a way to get the LAPACK qr-factorization routine to work with scipy. Then by chance I found that it was already there, under 'scipy.linalg.flapack.dgeqrf' ! So I used it and the computation is now improved over octave ^^ (I did not benchmark but I would say 4~5 times faster) cheers, Evan

2 1

Changes in SVN scipy_core
by Travis Oliphant Dec. 21, 2005

Dec. 21, 2005

Hi Folks, It's great to see the configuration stuff getting better in scipy.distutils. Thanks to Pearu for all his hard work on scipy.distutils. I want to let you all know about the recent changes in the SVN version of scipy_core. As you may recall, I dramatically improved the way the data in an array can be understood by improving the PyArray_Descr structure and making it an object. As part of that improvement, it became clear that the NOTSWAPPED flag was an anomaly and shouldn't be a flag on the array itself. The byte-order of the data is a property of the data-type descriptor. This is especially clear when considering records which according to the array protocol can have some fields in one byte order and some in another. As a result, I've removed the NOTSWAPPED flag from the array object. The equivalent of arr.flags.notswapped is now arr.dtypedescr.isnative. In C, the macro PyArray_ISNOTSWAPPED(arr) is still available but it checks the data-type descriptor. All C-API and python calls that used to pass a swap paramter along with a data-type descriptor have the swap paramter deleted. These C-API changes mean you will need to fully rebuild (remove the build directory and build) both scipy_core and scipy if you check out the new SVN version. I realize that the pace of development has been rapid --- but I'm on a tight schedule ;-) Hopefully, this wil be the last relatively major change to the code base for a while. All tests pass for me for full scipy after these changes. As we all know, however, there still may be remaining issues. One issue, for example, is that a.copy() returns a copy of the data (with the same data-type descriptor so no change in the byte-order is done). There is a new method a.newbyteorder(<arg>) which returns the equivalent of a.view(a.dtypedescr.newbyteorder(<arg>)) which returns a new view of the data using a different byte-order description. Note that the newbyteorder method of a dtypedescr object returns a new copy of the dtypedescr object with the byte-orders swapped if <arg> is not given or forces the byteorder to a particular value if arg is given (and not arg=='|' which means ignore). All fields of the dtypedescr object are (recursively) updated with the same rule as well. Look what comes along for the ride because of these changes: a = ones((5,), '<i4') # define a little-endian array b = ones((5,), '>i4') # define a big-endian array a.tostring() '\x01\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00' b.tostring() '\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00\x01\x00\x00\x00\x01' You can use these byte-order flags anywhere a data-type descriptor (dtype parameter) is required. I have not tested all the possibilities, of course, so there may still be outstanding issues. Note, however, that if you work with arrays that are not in native-byte order some operations will generally be slower. Regards, -Travis

10 19

Mac installers for scipy, etc.
by Chris Fonnesbeck Dec. 21, 2005

Dec. 21, 2005

For interested OS X users, I have provided mpkg installers for scipy_core and scipy from svn, as well as matplotlib 0.85.1 (cvs) and numarray 1.5.1. These were all built on OS X 10.4 on Python 2.4.2. http://trichech.us/mac Enjoy. -- Chris Fonnesbeck Atlanta, GA

3 4

Install problem - Overwriting message when importing scipy
by Hugo Gamboa Dec. 21, 2005

Dec. 21, 2005

Hi to all, I decided to make the effort to use the svn version of scipy. After adding the requested libraries to my kubuntu box, the core/scipy install process gave some warnings but managed to finish. The problem appears when the import is done where the following messages appear: >> import scipy Overwriting test = <module 'scipy.test' from '/usr/lib/python2.4/site-packages/scipy/test/__init__.pyc'> with <bound method ScipyTest.test of <scipy.test.testing.ScipyTest instance at 0xb4eec58c>> Overwriting utils = <module 'scipy.base.utils' from '/usr/lib/python2.4/site-packages/scipy/base/utils.pyc'> with <module ' scipy.utils' from '/usr/lib/python2.4/site-packages/scipy/utils/__init__.pyc'> Overwriting fftpack = <module 'scipy.basic.fftpack' from '/usr/lib/python2.4/site-packages/scipy/basic/fftpack.pyc'> with <module ' scipy.fftpack' from '/usr/lib/python2.4/site-packages/scipy/fftpack/__init__.pyc'> Overwriting fft = <function fft at 0xb4eef41c> with <function fft at 0xb48d23e4> Overwriting ifft = <function inverse_fft at 0xb4eef454> with <function ifft at 0xb48d241c> Overwriting randn = <built-in method randn of mtrand.RandomState object at 0xb7e58190> with <function randn at 0xb48a59cc> Some code that used to work with scipy managed to give a segmentation fault. Since is my first scipy installation I may missing some very basic step. Have anyone experienced and solved a similar problem? Are more details needed? Thanks in advance for your help. Hugo Gamboa

5 16

question about scipy.polynomial.poly1d
by Michael Blazej Dec. 21, 2005

Dec. 21, 2005

Mike here, this is my first time writing something, usually I'm just reading it but by question was. Is there a way to get different variables into poly1d such as 'y' or 'lamda'. Thank you all, this is a great way for new people to learn about scipy and python.

4 5