Mailman 3 July 2001 - NumPy-Discussion

Matrix exponential
by Nils Wagner 25 Jul '01

25 Jul '01

Hi, Has anyone written some functions concerning the matrix exponential expm(A) ? I am looking for demos illustrating the use of Padé approximation, Taylor series approximation, and eigenvalues and eigenvectors, respectively, to compute the matrix exponential. Thanks in advance. Nils Reference : SIAM Review, Vol. 20 (1979) pp.801-836 Moler, van Loan "Nineteen dubious ways to compute the exponential of a matrix"

1 0

A question on vector and matrix norms
by Nils Wagner 25 Jul '01

25 Jul '01

Hi, I'm interested in efficient implementations of various vector and matrix norms. Please can anyone give me some short examples regarding the following norms for vectors : 1. Frobenius norm 2. L_1 norm is sum of absolute values 3. L_2 norm 4. L_\infty or maximum norm 5. p-norm and matrices: matrix spectral norm or matrix 2-norm Frobenius norm ||A||_F p-norms ||A||_p Thanks in advance. Nils

1 0

Re: [Numpy-discussion] largest matrix element
by Tim Hochberg 25 Jul '01

25 Jul '01

Hi Scott, et al. First off I goofed -- that last reply should have gone to the list as a whole. Second, mere seconds after I pushed send I realized one problem with it: 'flat' only works for contiguous arrays. Its probably safer to use ravel after all, unless it's known that the array is contiguous. On the other hand, it would be slightly more efficient to ravel the matrix after taking its absolute value since ravel potentially allocates a new array and an array of Floats is going to be half the size of an array of Complexes. But then we know the resulting array is going to be contiguous. So after rewriting and erasing the thing a few times I end up with: max(abs(a).flat) Which is almost the same as my first one, except that the flat has been moved and now it should work when the array starts out noncontiguous. -tim > Hi Tim, > > Looks good to me! ;) > > Scott > > Tim Hochberg wrote: > > > > ----- Original Message ----- > > From: "Scott Ransom" <ransom(a)cfa.harvard.edu> > > Cc: <nils.wagner(a)freenet.de>; <numpy-discussion(a)lists.sourceforge.net> > > Sent: Tuesday, July 24, 2001 12:24 PM > > Subject: Re: [Numpy-discussion] largest matrix element > > > > > Charles G Waldman wrote: > > > > > > > > nils.wagner(a)freenet.de writes: > > > > > Given a complex Matrix A(m,n) > > > > > How can I get the absolute value of the largest matrix element > > > > > very efficiently ? > > > > > > > > How about > > > > > > > > sqrt(max((asarray(A)*conjugate(asarray(A))).real)) > > > > > > I if you want the global maximum you need a ravel in there: > > > > > > sqrt(max(ravel((asarray(a)*conjugate(asarray(a))).real))) > > > > I must be missing something. What's wrong with: > > > > max(abs(a.flat)) ? > > > > -tim > > -- > Scott M. Ransom Address: Harvard-Smithsonian CfA > Phone: (617) 496-7908 60 Garden St. MS 10 > email: ransom(a)cfa.harvard.edu Cambridge, MA 02138 > GPG Fingerprint: 06A9 9553 78BE 16DB 407B FFCA 9BFA B6FF FFD3 2989

1 0

largest matrix element
by nils.wagner＠freenet.de 25 Jul '01

25 Jul '01

Given a complex Matrix A(m,n) How can I get the absolute value of the largest matrix element very efficiently ? Nils Die schönen Seiten des Urlaubs bei Libri.de: Sonnige Bücher portofrei bestellen: http://www.libri.de/content/urlaub.html

3 2

20.1.0 undefined symbol: PyArray_API
by Mike Romberg 21 Jul '01

21 Jul '01

I have written a python module in C which is implemented as a loadable shared library. My module uses Numeric array objects and includes arrayobject.h. With Numeric-19 this used to work fine. Now my module will not load because the symbol PyArray_API is not found. I located this fellow in the arrayobject header file: /* C API address pointer */ #if defined(NO_IMPORT) || defined(NO_IMPORT_UFUNC) extern void **PyUFunc_API; #else #if defined(PY_UFUNC_UNIQUE_SYMBOL) void **PyUFunc_API; #else static void **PyUFunc_API; #endif So, it seems (at first glance) that this symbol is either static or extern based on the NO_IMPORT macro. My question is, can I just define NO_IMPORT before including arrayobject.h or does Numeric python need to be rebuilt so that this symbol gets exported? Thanks, Mike Romberg (romberg(a)fsl.noaa.gov)

1 0

Re: [Numpy-discussion] Cygwin
by Jochen Küpper 21 Jul '01

21 Jul '01

>>>>> Paul F Dubois wrote on Tue, 17 Jul 2001 12:53:52 -0700: [patch to Src/fastumathmodule.c] Paul> This file is no longer in the distribution. So why was it changed today? > cvs log Src/fastumathmodule.c ,---- | RCS file: /cvsroot/numpy/Numerical/Src/fastumathmodule.c,v | Working file: Src/fastumathmodule.c | head: 1.2 | branch: | locks: strict | access list: | symbolic names: | keyword substitution: kv | total revisions: 2; selected revisions: 2 | description: | ---------------------------- | revision 1.2 | date: 2001/07/20 03:24:33; author: teoliphant; state: Exp; lines: +18 -0 | Added inverse hyperbolic functions. | ---------------------------- | revision 1.1 | date: 2001/07/16 23:19:23; author: teoliphant; state: Exp; | Added fastumathmodule which doesn't emit exceptions. | ============================================================================= `---- To get that straight: I am looking at the Repository "Numerical" at :pserver:anonymous@cvs.numpy.sourceforge.net:/cvsroot/numpy I hope that is the correct module for an uptodate version of NumPy. If so, I would suggest the following patch to have it link successfully on Cygwin: Index: Src/fastumathmodule.c =================================================================== RCS file: /cvsroot/numpy/Numerical/Src/fastumathmodule.c,v retrieving revision 1.2 diff -u -r1.2 fastumathmodule.c --- Src/fastumathmodule.c 2001/07/20 03:24:33 1.2 +++ Src/fastumathmodule.c 2001/07/20 20:01:33 @@ -2141,7 +2141,7 @@ {NULL, NULL, 0} /* sentinel */ }; -void initfastumath() { +DL_EXPORT(void) initfastumath() { PyObject *m, *d, *s; /* Create the module and add the functions */ Thanks, Jochen -- University of North Carolina phone: 919-962-4403 Department of Chemistry phone: 919-962-1579 Venable Hall CB#3290 fax: 919-843-6041 Chapel Hill, NC 27599, USA GnuPG key: 44BCCD8E

1 0

Release 20.1
by Paul F. Dubois 21 Jul '01

21 Jul '01

Release 20.1 is available for your dining pleasure.

4 4

suggested doc improvement: C API, fortran arrays
by Peter Bienstman 20 Jul '01

20 Jul '01

Hi, In chapter 14 of the docs (Writing a C extension to NumPy), the last section (a less simple example) deals with mixing Fortran and C code. Wouldn't it be useful to mention the topic of different storage orders there, and to reflect this in the code by swapping some stride parameters? Could avoid a lot of frustration... Peter ------------------------------------- Peter Bienstman Department of Information Technology INTEC/IMEC - Ghent University St.-Pietersnieuwstraat 41 B-9000 Gent - Belgium E-mail: Peter.Bienstman(a)rug.ac.be Tel: +32 9 264 3445 Fax: +32 9 264 3593 -------------------------------------

1 0

Re: [Numpy-discussion] Kronecker product
by Charles G Waldman 19 Jul '01

19 Jul '01

Nils Wagner writes: > > How can I rearrange the result of outerproduct to the result of > Kronecker product with numpy ? > def kron(a,b): o = outerproduct(a,b) o.shape = a.shape + b.shape return concatenate(concatenate(o, axis=1), axis=1)

2 1

Re: Problem running Numeric-20.1 installer on NT4,SP3
by Charles L. Bowman, Ph.D. 19 Jul '01

19 Jul '01

Hello, I have python2.1.1c1 in I:\python21. It's up and running OK. I downloaded Numeric-20.1.0.win32-py2.1.exe, placed it in I:\python21, opened up a DOS window and ran the installer. The desktop flickered *very* briefly. Import Numeric returns "No module named Numeric". I cannot find any files named numeric.py. Does the installer assume python2.1 is in C:\python21? If so, how can I pass the correct drive information to the installer? Thanks for your help. Charles Bowman

1 0