From hinsen@ibs.ibs.fr Sun Feb 2 14:40:22 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Sun, 2 Feb 1997 15:40:22 +0100
Subject: [PYTHON MATRIX-SIG] mean, add.reduce, and MLab.py
In-Reply-To: <01BC0F7F.2698C2A0@misha.lcs.mit.edu> (message from Jim Hugunin
on Fri, 31 Jan 1997 14:00:43 -0500)
Message-ID: <199702021440.PAA29259@lmspc2.ibs.fr>
> 2) I don't understand people's complaints about the mean function at all.
> In fact, getting this to work right for multi-dimensional arrays was one
> of the plethora of reasons that reduce uses axis 0 by default.
>
> Could someone give me an example of how "x - add.reduce(x)/len(x)" doesn't
> work as expected? I'd argue that if this isn't the result you're looking
> for, then you don't quite understand NumPy's model of an array (I could be
> wrong...)
I agree to some degree. If you find that it systematically something
else than you expect, you have probably not chosen the right Python
representation for your data objects. But of course there are
occasions when explicit axis specifications are necessary - otherwise
they wouldn't be possible!
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Tue Feb 4 18:49:35 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Tue, 4 Feb 1997 19:49:35 +0100
Subject: [PYTHON MATRIX-SIG] netCDF module
Message-ID: <199702041849.TAA06352@lmspc2.ibs.fr>
I'd just like to remind all users of my netCDF module that there have
been quite a few changes recently (i.e. bug fixes), so it's worth
getting a new copy if you haven't done so recently. The URL is still
http://www.yi.com/home/HinsenKonrad/netcdf.html
Konrad.
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From Dale_Bertrand@brown.edu Thu Feb 6 13:50:41 1997
From: Dale_Bertrand@brown.edu (Dale Bertrand)
Date: Thu, 06 Feb 1997 08:50:41 -0500
Subject: [PYTHON MATRIX-SIG] Numeric typecodes
Message-ID: <199702061350.IAA14466@golden.brown.edu>
I'm using a pre 1.0 version of Numeric which does seem to support all the
typecodes in v1.0. For the available types for Floats are 'd' and 'f'.
Would someone please tell me how many bytes these use to represent floating
point data?
-Dale
---
Dale Bertrand
--------
email: dale_bertrand@brown.edu
quote: "I dislike forwards and silly quotes"
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Thu Feb 6 14:41:15 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Thu, 6 Feb 1997 15:41:15 +0100
Subject: [PYTHON MATRIX-SIG] Numeric typecodes
In-Reply-To: <199702061350.IAA14466@golden.brown.edu> (message from Dale
Bertrand on Thu, 06 Feb 1997 08:50:41 -0500)
Message-ID: <199702061441.PAA15196@lmspc2.ibs.fr>
> I'm using a pre 1.0 version of Numeric which does seem to support all the
> typecodes in v1.0. For the available types for Floats are 'd' and 'f'.
> Would someone please tell me how many bytes these use to represent floating
> point data?
That depends on your C compiler. 'd' stands for C 'double' (i.e.
standard Python float precision), and 'f' stands for C 'float'.
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From x-aes@telelogic.se Fri Feb 7 09:21:53 1997
From: x-aes@telelogic.se (Andy Eskilsson)
Date: 07 Feb 1997 10:21:53 +0100
Subject: [PYTHON MATRIX-SIG] NumPy on dos..
Message-ID: <ksraitvumm.fsf@telelogic.se>
Just thought I should drop a line and say that I successfully compiled
and integrated the NumPy module into python 1.4 for dos, with the help
of djgpp, (D. J. Delories gcc port).
There were one quirk, the compiler died if I tried to compile the
zlapack_lite.c file with -O2 optimizations, but when I switched down
to -O it compiled fine.
Otherwise it worked smooth, I had just to follow the instructions
(well I couldn't do that with a symbolic link, so I copied the files
instead) and make.. (uhm well I have bash for dos, so I let configure
do the makefiles for me.)
/Andy
p.s. I am not on this list, just thought you might be interested.
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From Dale_Bertrand@brown.edu Sat Feb 8 01:14:56 1997
From: Dale_Bertrand@brown.edu (Dale Bertrand)
Date: Fri, 07 Feb 1997 20:14:56 -0500
Subject: [PYTHON MATRIX-SIG] Numeric typecodes
Message-ID: <199702080114.UAA01999@golden.brown.edu>
>> I'm using a pre 1.0 version of Numeric which does seem to support all the
>> typecodes in v1.0. For the available types for Floats are 'd' and 'f'.
>> Would someone please tell me how many bytes these use to represent floating
>> point data?
>
>Same as the underlying data they represent, i.e. for a float, 'f', on a
>sparc (which you're on ;) i believe it's 4 bytes. A double, 'd', is 8
>bytes.
Since I can't throw 2 byte Float into an array, thinking about manipulating
the two bytes into a 4 byte float by adding two bytes in the right places
for each value. Will this work? I've made a diagram below.
** **
** ** 2 byte float
** **
| |
| ---
| |
** ** ** **
** ** ** ** 4 byte float
** ** ** **
^ ^
| |
(added zero bytes)
---
Dale Bertrand
--------
email: dale_bertrand@brown.edu
quote: "I dislike forwards and silly quotes"
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From szummer@media.mit.edu Wed Feb 19 03:17:52 1997
From: szummer@media.mit.edu (Martin Szummer)
Date: Tue, 18 Feb 1997 22:17:52 -0500
Subject: [PYTHON MATRIX-SIG] 2D slices of 1D array
Message-ID: <9702190317.AA05985@arsenal.media.mit.edu>
The following slice of a 1D array surprised me:
>>> c = array([1,2,3,4,5])
>>> c
1 2 3 4 5
>>> c[1:3, :]
EmptyArray
I was expecting either 1,2,3 or an error.
Is there any rational for this?
The behavior is the same for higher dimensional arrays as well.
(This is using Numeric 1.06a and Python 1.4)
-- Martin Szummer
MIT
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Wed Feb 19 09:03:02 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Wed, 19 Feb 1997 10:03:02 +0100
Subject: [PYTHON MATRIX-SIG] 2D slices of 1D array
In-Reply-To: <9702190317.AA05985@arsenal.media.mit.edu> (message from Martin
Szummer on Tue, 18 Feb 1997 22:17:52 -0500)
Message-ID: <199702190903.KAA00286@lmspc2.ibs.fr>
> The following slice of a 1D array surprised me:
>
> >>> c = array([1,2,3,4,5])
> >>> c
> 1 2 3 4 5
> >>> c[1:3, :]
> EmptyArray
>
> I was expecting either 1,2,3 or an error.
> Is there any rational for this?
None that I can think of. This ought to be an error.
Konrad.
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Thu Feb 20 00:08:38 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Wed, 19 Feb 1997 19:08:38 -0500
Subject: [PYTHON MATRIX-SIG] 2D slices of 1D array
Message-ID: <01BC1E98.50DE75F0@misha.lcs.mit.edu>
This will now raise an IndexError exception as it should have all along.
Thanks for the bug - Jim
-----Original Message-----
From: Konrad Hinsen [SMTP:hinsen@ibs.ibs.fr]
> The following slice of a 1D array surprised me:
>
> >>> c = array([1,2,3,4,5])
> >>> c
> 1 2 3 4 5
> >>> c[1:3, :]
> EmptyArray
>
> I was expecting either 1,2,3 or an error.
> Is there any rational for this?
None that I can think of. This ought to be an error.
Konrad.
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From djc@lct.com Fri Feb 21 17:23:20 1997
From: djc@lct.com (Duncan Child)
Date: Fri, 21 Feb 97 11:23:20 CST
Subject: [PYTHON MATRIX-SIG] Numerical Recipes Book
Message-ID: <9702211723.AA28149@ lct.com>
Hi all,
Have the Press et al algorithms been hooked up yet?
I just want to test a quick correlation thing this morning and
it would be perfect if a 'NumRec' module was already available.
I also have a large amount of my own algorithms that might need
to be put into new modules so I would be interested to see how
a large extension like the Numerical Recipes algorithms would
be structured.
Thanks a lot,
Duncan
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Fri Feb 21 17:47:18 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Fri, 21 Feb 1997 18:47:18 +0100
Subject: [PYTHON MATRIX-SIG] Numerical Recipes Book
In-Reply-To: <9702211723.AA28149@ lct.com> (djc@lct.com)
Message-ID: <199702211747.SAA10832@lmspc2.ibs.fr>
> Have the Press et al algorithms been hooked up yet?
As far as I know no.
> I just want to test a quick correlation thing this morning and
> it would be perfect if a 'NumRec' module was already available.
I don't know what you exactly you need, but perhaps the fft stuff
that comes with NumPy is sufficient for calculating your correlations.
> I also have a large amount of my own algorithms that might need
> to be put into new modules so I would be interested to see how
> a large extension like the Numerical Recipes algorithms would
> be structured.
Look at the wrappers for LAPACK etc. as an example. Also check out
SWIG, there's a pointer to it on the Python Web site.
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hohn@math.utah.edu Fri Feb 21 17:52:10 1997
From: hohn@math.utah.edu (Michael Hohn)
Date: Fri, 21 Feb 1997 10:52:10 -0700 (MST)
Subject: [PYTHON MATRIX-SIG] Numerical Recipes Book
In-Reply-To: <9702211723.AA28149@ lct.com> (djc@lct.com)
Message-ID: <199702211752.KAA17439@alab06.math.utah.edu>
>> ...
>>
>> Date: Fri, 21 Feb 97 11:23:20 CST
>> From: djc@lct.com (Duncan Child)
>> Sender: owner-matrix-sig@python.org
>> Precedence: bulk
>>
>>
>> Hi all,
>>
>> Have the Press et al algorithms been hooked up yet?
>>
>> ...
A warning about Numerical Recipies may be in order. For a summary of
why NOT to use this software, have a look at
http://math.jpl.nasa.gov/nr/nr.html
The netlib repository also has freely available software, and it's
usually much higher quality. See
http://www.netlib.org
Cheers,
Mike
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Fri Feb 21 20:02:41 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Fri, 21 Feb 1997 15:02:41 -0500
Subject: [PYTHON MATRIX-SIG] Numeric Python Beta 1 Released!
Message-ID: <01BC2008.4BF62120@misha.lcs.mit.edu>
I kept saying it was going to happen any day now...today's that day.
Check out http://www.sls.lcs.mit.edu/~jjh/numpy for the first beta release
of NumPy. It includes a source distribution that will work under both
windows and unix, and an easy to install binary for Win NT/95. Grab it,
install it, and let me know what you think (feel free to comment on the web
pages as well...).
Incompatible changes from the a6 release:
binarysearch has been renamed to searchsorted
Things that still need work (and for which help is greatly appreciated!):
MLab.py does not work in this release
Unix installation instructions need work, more info for different OS's
Bug fixes (all known bugs are squashed in this release...)
David Ascher needs to update his tutorial ;-)
Talk Jack Jansen into sharing his Macintosh port with me ;-)
Extension modules need to be built
Simple C extension example
Please send all bug reports directly to me (hugunin@mit.edu) and any
questions about the system to the matrix-sig. I fear that I've lost many
of the comments sent to me over the last five months due to a period of
immense disorganization in my life. I'm now much better set up to collect
bug reports and feature suggestions, so if I haven't addressed your
comments in this latest release, please send them to me again.
As usual, enjoy! - Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From da@maigret.cog.brown.edu Fri Feb 21 22:43:45 1997
From: da@maigret.cog.brown.edu (David Ascher)
Date: Fri, 21 Feb 1997 17:43:45 -0500 (EST)
Subject: [PYTHON MATRIX-SIG] Numeric Python Beta 1 Released!
In-Reply-To: <01BC2008.4BF62120@misha.lcs.mit.edu>
Message-ID: <Pine.SGI.3.95q.970221174131.24920B-100000@maigret>
> Things that still need work (and for which help is greatly appreciated!):
..
> David Ascher needs to update his tutorial ;-)
I'll do that when I can, but it won't be for at least 10 days, as I'll be
in the Bay Area job hunting starting Sunday.
Folks should feel free to send me specific requests if they have any. I
have a big mailbox. Job offers are also welcome...
--da
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From szummer@media.mit.edu Sun Feb 23 01:33:18 1997
From: szummer@media.mit.edu (Martin Szummer)
Date: Sat, 22 Feb 1997 20:33:18 -0500
Subject: [PYTHON MATRIX-SIG] Numeric benchmark suite?
Message-ID: <9702230133.AA08394@arsenal.media.mit.edu>
Hi,
I noticed some speed discrepancies between Python on different
platforms (with otherwise comparable CPUs). (This concerns 1.0a6
under Linux, HPUX, IRIX, and Alpha).
I would like to do a more thorough speed test,
so I was wondering if anybody has a good benchmark suite?
Ideally the benchmark suite should break up the results for different
operations, such as matrixmultiply, addition, FFT, and also measure different
data types (Integer, float, double, complex), and matrix sizes.
I suspect the speed discrepancies depend on the compiler (cc vs. gcc)
and compilation options. Any recommendations for compilers/options?
-- Martin
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Sun Feb 23 09:33:24 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Sun, 23 Feb 1997 10:33:24 +0100
Subject: [PYTHON MATRIX-SIG] Updated installation instructions for NumPy 1.0b1
In-Reply-To: <01BC2008.4BF62120@misha.lcs.mit.edu> (message from Jim Hugunin
on Fri, 21 Feb 1997 15:02:41 -0500)
Message-ID: <199702230933.KAA17488@lmspc2.ibs.fr>
There are two ways to install the numeric extensions.
**Built in python source tree
1) Get the source distribution to python and configure it.
2) In the top-level directory of the source tree, unpack NumPy
> gunzip NumPy-1.0b1.tar.gz
> tar -xvmf NumPy-1.0b1.tar .
2a) unpack NumPyLib in the same directory using the same steps.
This is not necessary if you have the "real" numeric libraries, but
in that case you will need to change the Setup script as described below.
3a) Edit Modules/Setup from the python distribution to add the following lines:
#This must occur above *shared* if you are building dynamic extensions
# Jim Hugunin's numeric python extensions
#The core functionality of multidimensional arrays
multiarray -I../NumPy/Include ../NumPy/multiarraymodule.c ../NumPy/arrayobject.c ../NumPy/ufuncobject.c
#The rest of this can be built dynamically and occur below *shared*
# NumPy path
NUMPYPATH=:$(DESTLIB)/NumPy
#The basic math functions (+,-,...,log,...,greater_equal)
fast_umath -I../NumPy/Include ../NumPy/fast_umathmodule.c -lm
umath -I../NumPy/Include ../NumPy/umathmodule.c -lm
#If you have a real fftpack library, include it here.
fftpack -I../NumPy/Include -I../NumPy/NumPyLib ../NumPy/fftpackmodule.c ../NumPy/NumPyLib/fftpack.c
#Link with the real lapack library (and F77/I77/BLAS) if you have it
lapack_lite -I../NumPy/Include -I../NumPy/NumPyLib ../NumPy/lapack_litemodule.c ../NumPy/NumPyLib/dlapack_lite.c ../NumPy/NumPyLib/zlapack_lite.c ../NumPy/NumPyLib/blas_lite.c ../NumPy/NumPyLib/f2c_lite.c
#Link with the real ranlib if you have it
ranlib -I../NumPy/Include -I../NumPy/NumPyLib ../NumPy/ranlibmodule.c ../NumPy/NumPyLib/ranlib.c ../NumPy/NumPyLib/com.c ../NumPy/NumPyLib/linpack.c
3b) Locate the line starting with COREPYTHONPATH= in Modules/Setup. Add
"$(NUMPYPATH)" (without the quotes) at the end of this line
4) Edit the file Makefile.in in the top-level directory of the Python
distribution. Locate the line starting with LIBSUBDIRS=. Add
" NumPy" (without the quotes) at the end of this line.
5) Add a symbolic link to the Numeric python libraries
(In the top level python directory)
> cd Lib; ln -s ../NumPy/NumPy NumPy
6) Configure, build and install the python distribution (see the Python
installation instructions for details):
> configure
> make install
If you have already configured your Python distribution, it is
not necessary to redo this step if you make the modification described
in step 4) also to the file Makefile. You should, however, recompile
everything and install the new Python binary, which will have a
different default search path than the standard version:
> make clean
> make install
**Built as a clean extension
1) Unpack NumPy anywhere
2) In NumPy, type:
> make -f Makefile.pre.in Makefile
> make
3) Add this NumPy directory to your PYTHONPATH environment variable.
You should run the supplied test cases in order to see if things are
working on your system:
> cd NumPy/test
> python test_all.py
If you plan to install extensions (i.e. C modules) that need NumPy,
note that they will have to include the file NumPy/Include/arrayobject.h.
The best solution is to copy this file into the directory that contains
the other Python include files (i.e. normally /usr/local/include/python1.4).
Then you won't have to do anything special to compile other extensions
later.
Note that there is documentation (ugly but reasonably complete) in
doc.html
The distribution was created by executing the following:
tar -cvf NumPy-1.0b1.tar NumPy/INSTALL NumPy/Makefile.pre.in NumPy/Setup NumPy/makefile_nt.msc NumPy/numpy_nt.def NumPy/numpy.mak NumPy/numpy.mdp NumPy/html/*.html NumPy/*.c NumPy/*.h NumPy/Include/* NumPy/NumPy/*.py NumPy/test/test_*.py NumPy/Demo/*.py NumPy/Misc/*
tar -cvf NumPyLib-1.0b1.tar NumPy/NumPyLib/*.c NumPy/NumPyLib/*.h
System specific notes:
**HP/UX**
Some of the code uses ANSI features and some of the code is standard C. This makes the HP compilier unhappy. There are two solutions:
1) Use Gnu's free "gcc" compilier
2) Use the following command sequence for compilation:
make -k
make OPT="-Ae -O"
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Sun Feb 23 09:33:43 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Sun, 23 Feb 1997 10:33:43 +0100
Subject: [PYTHON MATRIX-SIG] Fixed version of LinearAlgebra.py
In-Reply-To: <01BC2008.4BF62120@misha.lcs.mit.edu> (message from Jim Hugunin
on Fri, 21 Feb 1997 15:02:41 -0500)
Message-ID: <199702230933.KAA17492@lmspc2.ibs.fr>
# This module is a lite version of LinAlg.py module which contains
# high-level Python interface to the LAPACK library. The lite version
# only accesses the following LAPACK functions: dgesv, zgesv, dgeev,
# zgeev, dgesvd, zdesvd, dgelss, zgelss.
import Numeric
import copy
import lapack_lite
# Error object
LinAlgError = 'LinearAlgebraError'
# Helper routines
_lapack_type = {'f': 0, 'd': 1, 'F': 2, 'D': 3}
_lapack_letter = ['s', 'd', 'c', 'z']
_array_kind = {'i':0, 'l': 0, 'f': 0, 'd': 0, 'F': 1, 'D': 1}
_array_precision = {'i': 1, 'l': 1, 'f': 0, 'd': 1, 'F': 0, 'D': 1}
_array_type = [['f', 'd'], ['F', 'D']]
def _commonType(*arrays):
kind = 0
# precision = 0
# force higher precision in lite version
precision = 1
for a in arrays:
t = a.typecode()
kind = max(kind, _array_kind[t])
precision = max(precision, _array_precision[t])
return _array_type[kind][precision]
def _castCopyAndTranspose(type, *arrays):
cast_arrays = ()
for a in arrays:
if a.typecode() == type:
cast_arrays = cast_arrays + (copy.copy(Numeric.transpose(a)),)
else:
cast_arrays = cast_arrays + (copy.copy(
Numeric.transpose(a).astype(type)),)
if len(cast_arrays) == 1:
return cast_arrays[0]
else:
return cast_arrays
def _assertRank2(*arrays):
for a in arrays:
if len(a.shape) != 2:
raise LinAlgError, 'Array must be two-dimensional'
def _assertSquareness(*arrays):
for a in arrays:
if max(a.shape) != min(a.shape):
raise LinAlgError, 'Array must be square'
# Linear equations
def solve_linear_equations(a, b):
one_eq = len(b.shape) == 1
if one_eq:
b = b[:, Numeric.NewAxis]
_assertRank2(a, b)
_assertSquareness(a)
n_eq = a.shape[0]
n_rhs = b.shape[1]
if n_eq != b.shape[0]:
raise LinAlgError, 'Incompatible dimensions'
t =_commonType(a, b)
# lapack_routine = _findLapackRoutine('gesv', t)
if _array_kind[t] == 1: # Complex routines take different arguments
lapack_routine = lapack_lite.zgesv
else:
lapack_routine = lapack_lite.dgesv
a, b = _castCopyAndTranspose(t, a, b)
pivots = Numeric.zeros(n_eq, 'l')
results = lapack_routine(n_eq, n_rhs, a, n_eq, pivots, b, n_eq, 0)
if results['info'] > 0:
raise LinAlgError, 'Singular matrix'
if one_eq:
return copy.copy(Numeric.ravel(b))
else:
return copy.copy(Numeric.transpose(b))
# Matrix inversion
def inverse(a):
return solve_linear_equations(a, Numeric.identity(a.shape[0]))
# Eigenvalues
def eigenvalues(a):
_assertRank2(a)
_assertSquareness(a)
t =_commonType(a)
real_t = _array_type[0][_array_precision[t]]
a = _castCopyAndTranspose(t, a)
n = a.shape[0]
dummy = Numeric.zeros((1,), t)
lwork = max(1,6*n) # minimum value: max(1,3*n) real, max(1,2*n) complex
work = Numeric.zeros((lwork,), t)
if _array_kind[t] == 1: # Complex routines take different arguments
lapack_routine = lapack_lite.zgeev
w = Numeric.zeros((n,), t)
rwork = Numeric.zeros((n,),real_t)
results = lapack_routine('N', 'N', n, a, n, w,
dummy, 1, dummy, 1, work, lwork, rwork, 0)
else:
lapack_routine = lapack_lite.dgeev
wr = Numeric.zeros((n,), t)
wi = Numeric.zeros((n,), t)
results = lapack_routine('N', 'N', n, a, n, wr, wi,
dummy, 1, dummy, 1, work, lwork, 0)
if Numeric.logical_and.reduce(Numeric.equal(wi, 0.)):
w = wr
else:
w = wr+1j*wi
if results['info'] > 0:
raise LinAlgError, 'Eigenvalues did not converge'
return w
# Eigenvectors
def eigenvectors(a):
"""eigenvectors(a) returns u,v where u is the eigenvalues and
v is a matrix of eigenvectors with vector v[i] corresponds to
eigenvalue u[i]. Satisfies the equation matrixmultiply(a, v[i]) = u[i]*v[i]
"""
_assertRank2(a)
_assertSquareness(a)
t =_commonType(a)
real_t = _array_type[0][_array_precision[t]]
a = _castCopyAndTranspose(t, a)
n = a.shape[0]
lwork = max(1,8*n) # minimum value: max(1,4*n) real, max(1,2*n) complex
work = Numeric.zeros((lwork,), t)
dummy = Numeric.zeros((1,), t)
if _array_kind[t] == 1: # Complex routines take different arguments
lapack_routine = lapack_lite.zgeev
w = Numeric.zeros((n,), t)
rwork = Numeric.zeros((lwork,),real_t)
v = Numeric.zeros((n,n), t)
results = lapack_routine('N', 'V', n, a, n, w,
dummy, 1, v, n, work, lwork, rwork, 0)
else:
lapack_routine = lapack_lite.dgeev
wr = Numeric.zeros((n,), t)
wi = Numeric.zeros((n,), t)
vr = Numeric.zeros((n,n), t)
results = lapack_routine('N', 'V', n, a, n, wr, wi,
dummy, 1, vr, n, work, lwork, 0)
if Numeric.logical_and.reduce(Numeric.equal(wi, 0.)):
w = wr
v = vr
else:
w = wr+1j*wi
v = Numeric.array(vr,Numeric.Complex)
ind = Numeric.nonzero(
Numeric.equal(
Numeric.equal(wi,0.0) # true for real e-vals
,0) # true for complex e-vals
) # indices of complex e-vals
for i in range(len(ind)/2):
v[ind[2*i]] = vr[ind[2*i]] + 1j*vr[ind[2*i+1]]
v[ind[2*i+1]] = vr[ind[2*i]] - 1j*vr[ind[2*i+1]]
if results['info'] > 0:
raise LinAlgError, 'Eigenvalues did not converge'
return w,v
# Singular value decomposition
def singular_value_decomposition(a):
_assertRank2(a)
n = a.shape[1]
m = a.shape[0]
t =_commonType(a)
real_t = _array_type[0][_array_precision[t]]
a = _castCopyAndTranspose(t, a)
s = Numeric.zeros((min(n,m),), real_t)
u = Numeric.zeros((m, m), t)
lwork = 10*max(m,n) # minimum value: max(3*min(m,n)+max(m,n),5*min(m,n)-4)
work = Numeric.zeros((lwork,), t)
vt = Numeric.zeros((n, n), t)
if _array_kind[t] == 1: # Complex routines take different arguments
lapack_routine = lapack_lite.zgesvd
rwork = Numeric.zeros((max(3*min(m,n),5*min(m,n)-4),), real_t)
results = lapack_routine('A', 'A', m, n, a, m, s, u, m, vt, n,
work, lwork, rwork, 0)
else:
lapack_routine = lapack_lite.dgesvd
results = lapack_routine('A', 'A', m, n, a, m, s, u, m, vt, n,
work, lwork, 0)
if results['info'] > 0:
raise LinAlgError, 'SVD did not converge'
return copy.copy(Numeric.transpose(u)), s, copy.copy(Numeric.transpose(vt))
# Generalized inverse
def generalized_inverse(a):
u, s, vt = singular_value_decomposition(a)
m = u.shape[0]
n = vt.shape[0]
cutoff = 1.e-10*Numeric.maximum.reduce(s)
sm = Numeric.zeros((n, m), s.typecode())
for i in range(min(n,m)):
if s[i] > cutoff:
sm[i,i] = 1./s[i]
return Numeric.matrixmultiply(Numeric.matrixmultiply(Numeric.transpose(vt),
sm),
Numeric.transpose(u))
# Determinant
def determinant(a):
_assertRank2(a)
_assertSquareness(a)
ev = eigenvalues(a)
return Numeric.product(ev)
# Linear Least Squares
def linear_least_squares(a, b, rcond=1.e-10):
"""solveLinearLeastSquares(a,b) returns x,resids,rank,s
where x minimizes 2-norm(|b - Ax|)
resids is the sum square residuals
rank is the rank of A
s is an rank of the singual values of A in desending order
If b is a matrix then x is also a matrix with corresponding columns.
If the rank of A is less than the number of columns of A or greater than
the numer of rows, then residuals will be returned as an empty array
otherwise resids = sum((b-dot(A,x)**2).
Singular values less than s[0]*rcond are treated as zero.
"""
one_eq = len(b.shape) == 1
if one_eq:
b = b[:, Numeric.NewAxis]
_assertRank2(a, b)
m = a.shape[0]
n = a.shape[1]
n_rhs = b.shape[1]
ldb = max(n,m)
if m != b.shape[0]:
raise LinAlgError, 'Incompatible dimensions'
t =_commonType(a, b)
real_t = _array_type[0][_array_precision[t]]
bstar = Numeric.zeros((ldb,n_rhs),t)
bstar[:b.shape[0],:n_rhs] = copy.copy(b)
a,bstar = _castCopyAndTranspose(t, a, bstar)
lwork = 8*min(n,m) + max([2*min(m,n),max(m,n),n_rhs])
s = Numeric.zeros((min(m,n),),real_t)
work = Numeric.zeros((lwork,), t)
if _array_kind[t] == 1: # Complex routines take different arguments
lapack_routine = lapack_lite.zgelss
rwork = Numeric.zeros((5*min(m,n)-1,), real_t)
results = lapack_routine( m, n, n_rhs, a, m, bstar,ldb , s, rcond,
0,work,lwork,rwork,0 )
else:
lapack_routine = lapack_lite.dgelss
results = lapack_routine( m, n, n_rhs, a, m, bstar,ldb , s, rcond,
0,work,lwork,0 )
if results['info'] > 0:
raise LinAlgError, 'SVD did not converge in Linear Least Squares'
resids = Numeric.array([],t)
if one_eq:
x = copy.copy(Numeric.ravel(bstar)[:n])
if (results['rank']==n) and (m>n):
resids = Numeric.array([Numeric.sum((Numeric.ravel(bstar)[n:])**2)])
else:
x = copy.copy(Numeric.transpose(bstar)[:n,:])
if (results['rank']==n) and (m>n):
resids = copy.copy(Numeric.sum((Numeric.transpose(bstar)[n:,:])**2))
return x,resids,results['rank'],copy.copy(s[:min(n,m)])
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Sun Feb 23 09:35:08 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Sun, 23 Feb 1997 10:35:08 +0100
Subject: [PYTHON MATRIX-SIG] Elementary test suite for LinearAlgebra
In-Reply-To: <01BC2008.4BF62120@misha.lcs.mit.edu> (message from Jim Hugunin
on Fri, 21 Feb 1997 15:02:41 -0500)
Message-ID: <199702230935.KAA17500@lmspc2.ibs.fr>
from Numeric import *
from LinearAlgebra import *
def test(a, b):
print "All numbers printed should be (almost) zero:"
x = solve_linear_equations(a, b)
check = b - matrixmultiply(a, x)
print check
a_inv = inverse(a)
check = matrixmultiply(a, a_inv)-identity(a.shape[0])
print check
ev = eigenvalues(a)
evalues, evectors = eigenvectors(a)
check = ev-evalues
print check
evectors = transpose(evectors)
check = matrixmultiply(a, evectors)-evectors*evalues
print check
u, s, vt = singular_value_decomposition(a)
check = a - Numeric.matrixmultiply(u*s, vt)
print check
a_ginv = generalized_inverse(a)
check = matrixmultiply(a, a_ginv)-identity(a.shape[0])
print check
det = determinant(a)
check = det-multiply.reduce(evalues)
print check
x, residuals, rank, sv = linear_least_squares(a, b)
check = b - matrixmultiply(a, x)
print check
print rank-a.shape[0]
print sv-s
a = array([[1.,2.], [3.,4.]])
b = array([2., 1.])
test(a, b)
a = a+0j
b = b+0j
test(a, b)
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Sun Feb 23 09:32:55 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Sun, 23 Feb 1997 10:32:55 +0100
Subject: [PYTHON MATRIX-SIG] Numeric Python Beta 1 Released!
In-Reply-To: <01BC2008.4BF62120@misha.lcs.mit.edu> (message from Jim Hugunin
on Fri, 21 Feb 1997 15:02:41 -0500)
Message-ID: <199702230932.KAA17482@lmspc2.ibs.fr>
> I kept saying it was going to happen any day now...today's that day.
And it partly ruined my day, since of course I couldn't resist installing
and testing the new version immediately, although I had only come to
start a calculation job :-(
So far for the first complaint ;-)
> Check out http://www.sls.lcs.mit.edu/~jjh/numpy for the first beta release
> of NumPy. It includes a source distribution that will work under both
> windows and unix, and an easy to install binary for Win NT/95. Grab it,
> install it, and let me know what you think (feel free to comment on the web
> pages as well...).
The Web page is fine, I didn't expect such a luxury! Maybe you should add
pointers to code based on NumPy; I'll start immodestly by proposing my
own:
http://yi.com/home/HinsenKonrad/python.html
http://yi.com/home/HinsenKonrad/netcdf.html
Other stuff that comes to mind is the full LAPACK interface and the Gist
interface.
> Unix installation instructions need work, more info for different OS's
I corrected the installation instructions for the first choice
(installation in the Python source tree); I'll send the new version
around to everyone so that others can check whether it's OK for them.
I did not try the second choice, but I strongly suspect that the
instructions are incomplete. The shared libraries will end up in
NumPy, whereas the Python modules are in NumPy/NumPy. So *both*
directories must be added to PYTHONPATH, or one or the other set of
files must be copied. Personally I'd prefer to copy some files rather
than mess up my PYTHONPATH needlessly, but I'd rather let the people
who use it decide.
I remember some reports about problems with the second installation
method on some machines due to the shared library libnumpymodule.
That ought to be mentioned somewhere with a list of affected machines.
> Bug fixes (all known bugs are squashed in this release...)
I found only one trivial bug in the new release: test/test_all.py
fails under Unix with a syntax error because of the CRs in the file
test_items.py. They should be removed.
There have been a couple of bugs in LinearAlgebra.py for a while,
which I have corrected in the course of time, but never sent to
anyone. I have now done some superficial test of all functions, and
I'll send the fixed file plus my "test suite" around to everyone - it
would be good to have this tested on other machines as well.
I'll try to improve some functions later, but that won't change the
interface, so it's not urgent.
> questions about the system to the matrix-sig. I fear that I've lost many
> of the comments sent to me over the last five months due to a period of
> immense disorganization in my life. I'm now much better set up to collect
> bug reports and feature suggestions, so if I haven't addressed your
> comments in this latest release, please send them to me again.
I still think that it would be better to give Numeric.dot() the
functionality of Numeric.matrixmultiply() and remove the latter (or
keep it for compatibility, but not document it). There is no need for
two functions, and many people will probably use dot() although what
they need is matrixmultiply().
I also still think that the generalized version of diagonal() and trace()
that I posted a while ago would be more suitable than the current
version that works only on rank-2 arrays. Here's my version again:
------------------------------------------------------------
def diagonal(a, axis1=0, axis2=1, offset=0):
a = array(a)
if axis2 < axis1: axis1, axis2 = axis2, axis1
if axis2 > 1:
new_axes = range(len(a.shape))
del new_axes[axis2]; del new_axes[axis1]
new_axes[0:0] = [axis1, axis2]
a = transpose(a, new_axes)
s = a.shape
n1 = s[0]
n2 = s[1]
n = n1*n2
s = (n,) + s[2:]
a = reshape(a, s)
if offset < 0: offset = n2-offset-1
return take(a, arange(offset, n, n2+1), 0)
def trace(a, axis1=0, axis2=1, offset=0):
return add.reduce(diagonal(a, axis1, axis2, offset))
------------------------------------------------------------
That's it for now, more complaints later ;-)
Konrad.
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From acannon@geog.ubc.ca Sun Feb 23 23:40:27 1997
From: acannon@geog.ubc.ca (Alex Cannon)
Date: Sun, 23 Feb 1997 15:40:27 -0800 (PST)
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 install problem
In-Reply-To: <199702230933.KAA17488@lmspc2.ibs.fr>
Message-ID: <Pine.GSO.3.95.970223153434.11325B-100000@dragon>
I've just downloaded and executed the NumPy 1.0b1 binary for Win95/NT and
am getting:
!
Please run this installation
in a DOS box. It doesn't appear that a setup executable etc. is being
extracted. Has anyone else experienced a similar problem? This is with
Win95 4.00.950a.
Alex
------------------------------------------------------------
Alex Cannon (acannon@geog.ubc.ca) Geography 240A
Atmospheric Science Programme Tel: (604) 822-2269
University of British Columbia, B.C. Fax: (604) 822-6150
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From amullhau@ix.netcom.com Mon Feb 24 00:09:44 1997
From: amullhau@ix.netcom.com (Andrew P. Mullhaupt)
Date: Sun, 23 Feb 1997 19:09:44 -0500
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 install problem
References: <Pine.GSO.3.95.970223153434.11325B-100000@dragon>
Message-ID: <3310DC48.724A@ix.netcom.com>
Alex Cannon wrote:
>
> I've just downloaded and executed the NumPy 1.0b1 binary for Win95/NT and
> am getting:
>
> !
> Please run this installation
>
> in a DOS box. It doesn't appear that a setup executable etc. is being
> extracted. Has anyone else experienced a similar problem? This is with
> Win95 4.00.950a.
I have the same. However, at least PythonWin no longer crashes all the
time.
Later,
Andrew Mullhaupt
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From ta.short@pti-us.com Mon Feb 24 16:16:42 1997
From: ta.short@pti-us.com (Tom Short)
Date: Mon, 24 Feb 1997 11:16:42 -0500 (EST)
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 install problem
Message-ID: <3.0.1.16.19970224111628.2bdf4e9a@mailhost1473>
At 07:09 PM 2/23/97 -0500, you wrote:
>Alex Cannon wrote:
>>
>> I've just downloaded and executed the NumPy 1.0b1 binary for Win95/NT and
>> am getting:
>>
>> !
>> Please run this installation
>>
>> in a DOS box. It doesn't appear that a setup executable etc. is being
>> extracted. Has anyone else experienced a similar problem? This is with
>> Win95 4.00.950a.
>
I got this too but then noticed that the file size was not right (should be
408,830 bytes), so Netscape didn't download it correctly (I think it might
be trying to fetch it as text instead of binary). I used another web
download tool (webfetch), and it worked okay.
- Tom
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From barton@simsg1.mdc.com Mon Feb 24 16:36:01 1997
From: barton@simsg1.mdc.com (Brien Barton)
Date: Mon, 24 Feb 1997 08:36:01 -0800
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 install problem
References: <Pine.GSO.3.95.970223153434.11325B-100000@dragon> <3310DC48.724A@ix.netcom.com>
Message-ID: <3311C371.41C6@simsg1.mdc.com>
Andrew P. Mullhaupt wrote:
>
> Alex Cannon wrote:
> >
> > I've just downloaded and executed the NumPy 1.0b1 binary for Win95/NT and
> > am getting:
> >
> > !
> > Please run this installation
> >
> > in a DOS box. It doesn't appear that a setup executable etc. is being
> > extracted. Has anyone else experienced a similar problem? This is with
> > Win95 4.00.950a.
I executed the NumPy installation binary by double clicking on it from
Explorer and it installed fine.
-- Brien Barton
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Mon Feb 24 17:05:42 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Mon, 24 Feb 1997 12:05:42 -0500
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 install problem
Message-ID: <01BC224B.0F6155A0@misha.lcs.mit.edu>
I just changed the file name on the web page from NumPy-10-b1.EXE to
NumPy-10-b1.exe. This was the only thing I could see about the link that
might have been causing Netscape confusion over the file's format. Those
of you who have had problems should try downloading from this new link and
see if it fixes things.
As mentioned before, the file should have a size of 408,830 bytes when
downloaded to your system. If this is not the case, then there's something
wrong with your transfer.
Let me know if this works/fails - Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Mon Feb 24 18:22:59 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Mon, 24 Feb 1997 19:22:59 +0100
Subject: [PYTHON MATRIX-SIG] HP problem
Message-ID: <199702241822.TAA22379@lmspc2.ibs.fr>
Just in case anyone else tries the new NumPy under HP/UX 9.01: it seems
the HP compiler can't handle dlapack_lite.c and zlapack_lite.c; it
crashes with an "internal error" and asks me to contact HP support.
There may of course be some patch that fixes this, but the easiest
solution is to use gcc, which works fine, or to use the Fortran
LAPACK, which is probably the best solution.
Konrad.
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From susan@pcmdi.llnl.gov Mon Feb 24 18:53:43 1997
From: susan@pcmdi.llnl.gov (Susan Peterson)
Date: Mon, 24 Feb 1997 10:53:43 -0800
Subject: [PYTHON MATRIX-SIG] Installing New Numerics
Message-ID: <9702241053.ZM29541@penguin.llnl.gov>
Hi,
I've installed the new Numerics, but the test program isn't running.
After removing the CRs in the test_items.py file, when I run the
test_all.py I get the following error (after it has run for awhile)
Traceback (innermost last):
File "test_all.py", line 48, in ?
test_lines(fp.readlines())
File "test_all.py", line 29, in test_lines
out = eval(line)
File "<string>", line 0, in ?
NameError: searchsorted
Any ideas what I should do?
Please reply to susan@pcmdi.llnl.gov
Thanks in advance for your help.
-Susan
--
Susan Peterson
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From susan@pcmdi.llnl.gov Mon Feb 24 18:55:43 1997
From: susan@pcmdi.llnl.gov (Susan Peterson)
Date: Mon, 24 Feb 1997 10:55:43 -0800
Subject: No subject
Message-ID: <9702241055.ZM29560@penguin.llnl.gov>
Hi,
I've installed the new Numerics, but the test program isn't running.
After removing the CRs in the test_items.py file, when I run the
test_all.py I get the following error (after it has run for awhile)
Traceback (innermost last):
File "test_all.py", line 48, in ?
test_lines(fp.readlines())
File "test_all.py", line 29, in test_lines
out = eval(line)
File "<string>", line 0, in ?
NameError: searchsorted
Any ideas what I should do?
Please reply to susan@pcmdi.llnl.gov
Thanks in advance for your help.
-Susan
--
Susan Peterson
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From furnish@laura.llnl.gov Mon Feb 24 20:23:31 1997
From: furnish@laura.llnl.gov (Geoffrey Furnish)
Date: Mon, 24 Feb 1997 12:23:31 -0800
Subject: [PYTHON MATRIX-SIG] HP problem
In-Reply-To: <199702241822.TAA22379@lmspc2.ibs.fr>
References: <199702241822.TAA22379@lmspc2.ibs.fr>
Message-ID: <199702242026.PAA10383@python.org>
Konrad Hinsen writes:
> Just in case anyone else tries the new NumPy under HP/UX 9.01: it seems
> the HP compiler can't handle dlapack_lite.c and zlapack_lite.c; it
> crashes with an "internal error" and asks me to contact HP support.
> There may of course be some patch that fixes this, but the easiest
> solution is to use gcc, which works fine, or to use the Fortran
> LAPACK, which is probably the best solution.
Using gcc is not really fine except under certain conditions. If you
try to compile with -g, you can't make a shared module b/c of problems
with the size of some symbol table (and using -fPIC doesn't solve the
problem).
Exactly /how/ does one configure out the lapack_lite stuff and switch
to using the fortran versions. That has not been convered in anything
I've personally read.
--
Geoffrey Furnish email: furnish@llnl.gov
LLNL X/ICF phone: 510-424-4227 fax: 510-423-0925
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Mon Feb 24 21:29:21 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Mon, 24 Feb 1997 16:29:21 -0500
Subject: [PYTHON MATRIX-SIG] RE: NameError: searchsorted
Message-ID: <01BC226F.E47B4790@misha.lcs.mit.edu>
Susan Peterson writes:
> Hi,
>
> I've installed the new Numerics, but the test program isn't running.
> After removing the CRs in the test_items.py file, when I run the
> test_all.py I get the following error (after it has run for awhile)
>
> Traceback (innermost last):
> File "test_all.py", line 48, in ?
> test_lines(fp.readlines())
> File "test_all.py", line 29, in test_lines
> out = eval(line)
> File "<string>", line 0, in ?
> NameError: searchsorted
>
> Any ideas what I should do?
This error indicates that you are referencing an old version of Numeric.py.
This could either be due to a path problem pointing to an old
installation, or due to an old .pyc file being used instead of the new .py.
Hope this helps - Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From barton@simsg1.mdc.com Mon Feb 24 23:53:08 1997
From: barton@simsg1.mdc.com (Brien Barton)
Date: Mon, 24 Feb 1997 15:53:08 -0800 (PST)
Subject: [PYTHON MATRIX-SIG] NumPy test failing
Message-ID: <Pine.SGI.3.95.970224154159.10491B-100000@simsg1.mdc.com>
I installed the new NumPy 1.0b1 on an SGI system and tried to run the
test_all.py script. It gets an error on the test that tries to verify that,
if a is an array, then array(a) returns a copy of a. The lines from
test_items.py are:
b = array(a)
b[0] = 66
a[0]
>0
This test is apparently trying to verify that after the line "b = array(a)",
b and a are two different objects, but on my system they are the same.
Anyone else seeing this?
-- Brien
===============================================================================
Brien Barton ISSC Corp, c/o McDonnell Douglas Space & Defense Systems
Huntington Beach, CA
email: barton@simsg1.mdc.com, voice: (714)896-2249, fax:(714)896-5939
"My witty proclivities are nothing compared to my ludicrous ineptitudes."
- Bob Hope
===============================================================================
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From amullhau@ix.netcom.com Tue Feb 25 03:46:59 1997
From: amullhau@ix.netcom.com (Andrew P. Mullhaupt)
Date: Mon, 24 Feb 1997 22:46:59 -0500
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 install problem
References: <01BC224B.0F6155A0@misha.lcs.mit.edu>
Message-ID: <331260B3.7B5E@ix.netcom.com>
Jim Hugunin wrote:
>
> As mentioned before, the file should have a size of 408,830 bytes when
> downloaded to your system. If this is not the case, then there's something
> wrong with your transfer.
God only knows what is going on here. I tried again, and the download
put the file in NumPy-10-b1.txt. It wsa the right size, though, and
after renaming, it ran, claiming to have installed NumPy. PythonWin
doesn't seem to be all that numeric yet, since
resize(0, [2, 2])
results in a NameError.
Maybe after rebooting this will be better.
Later,
Andrew Mullhaupt
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From jhauser@ifm.uni-kiel.de Tue Feb 25 09:06:48 1997
From: jhauser@ifm.uni-kiel.de (janko hauser)
Date: Tue, 25 Feb 1997 10:06:48 +0100
Subject: [PYTHON MATRIX-SIG] inconsistency in array()?
Message-ID: <9702250906.AA08749@flores.ifm.uni-kiel.de>
Hello, I have tried the new NumPy-beta. I first had some problems with
the path, I mixed up the old and the new distribution. I hope that the
next points are unaffected by this :-)
I think the asarray() function is broken, and with this the shape()
function too. In Numeric.py (+244) array() is called with a keyword,
wich it doesn't understand. I had a look in the old and new
multiarray-code and the former method asarray has changed to a Numeric
function. Is this right? And why, may I ask?
UserArray is broken but I append a version wich hopefully is ok. (It
is broken, because it is not possible to use less() or something like
that) There is one remaining problem, because I can't make a
reference to a slice of the UserArray-object, I only get copys, wich
is suboptimal. Any suggestions? :-)
The benefit of the new implementation is, that I can use all functions
with an instance of this class, without definig them in the
UserArray-class.
__Janko
Here comes a new UserArray: (If there are bugs, please shout loud, I
need this class)
## --------------- cut here ----------------------------------------
#!/usr/local/bin/python
from Numeric import *
import string
class UserArray:
def __init__(self, data, typecode = None):
# Needs more testing
if typecode == None:
self.array = asarray(data)
else:
self.array = array(data, typecode)
self.shape = self.array.shape
self.typecode = self.array.typecode()
self.name = string.split(str(self.__class__))[1]
def __repr__(self):
return self.name+"\n"+str(self.array)
def __array__(self,t=None):
if t: return asarray(self.array,t)
return asarray(self.array)
def __float__(self):
return float(asarray(self.array))
# Array as sequence
def __len__(self): return len(self.array)
def __getitem__(self, index):
return self._rc(self.array[index])
def __getslice__(self, i, j):
return self._rc(self.array[i:j])
def __setitem__(self, index, value): self.array[index] = asarray(value,self.typecode)
def __setslice__(self, i, j, value): self.array[i:j] = asarray(value)
def __del__(self):
for att in dir(self):
delattr(self,att)
#del(self)
def __abs__(self): return self._rc(absolute(self.array))
def __neg__(self): return self._rc(-self.array)
def __add__(self, other):
return self._rc(self.array+asarray(other))
__radd__ = __add__
def __sub__(self, other):
return self._rc(self.array-asarray(other))
def __rsub__(self, other):
return self._rc(asarray(other)-self.array)
def __mul__(self, other):
return self._rc(multiply(self.array,asarray(other)))
__rmul__ = __mul__
def __div__(self, other):
return self._rc(divide(self.array,asarray(other)))
def __rdiv__(self, other):
return self._rc(divide(asarray(other),self.array))
def __pow__(self,other):
return self._rc(power(self.array,asarray(other)))
def __rpow__(self,other):
return self._rc(power(asarray(other),self.array))
def __sqrt__(self):
return self._rc(sqrt(self.array))
def tostring(self): return self.array.tostring()
def byteswapped(self): return self._rc(self.array.byteswapped())
def asType(self, typecode): return self._rc(self.array.asType(typecode))
def typecode(self): return self.array.typecode()
def itemsize(self): return self.array.itemsize()
def isContiguous(self): return self.array.isContiguous()
UserArray._rc=UserArray
#############################################################
# Test of class UserArray
#############################################################
if __name__ == '__main__':
import Numeric
temp=reshape(arange(10000),(100,100))
ua=UserArray(temp)
# new object created begin test
print dir(ua)
print shape(ua),ua.shape # I have changed Numeric.py
ua_small=ua[:3,:5]
print ua_small
ua_small[0,0]=10 # this did not change ua[0,0], wich is not normal behavior
print ua_small[0,0],ua[0,0]
print sin(ua_small)/3.*6.+sqrt(ua_small**2)
print less(ua_small,103),type(less(ua_small,103))
print type(ua_small*reshape(arange(15),shape(ua_small)))
print reshape(ua_small,(5,3))
print transpose(ua_small)
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Tue Feb 25 09:24:45 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Tue, 25 Feb 1997 10:24:45 +0100
Subject: [PYTHON MATRIX-SIG] HP problem
In-Reply-To: <199702242026.PAA10383@python.org> (message from Geoffrey Furnish
on Mon, 24 Feb 1997 12:23:31 -0800)
Message-ID: <199702250924.KAA24804@lmspc2.ibs.fr>
> Using gcc is not really fine except under certain conditions. If you
> try to compile with -g, you can't make a shared module b/c of problems
> with the size of some symbol table (and using -fPIC doesn't solve the
> problem).
True, but I see no reason to compile the LAPACK code with -g. I use
gcc only for the two modules that cc can't handle, so that shouldn't
cause too much trouble.
> Exactly /how/ does one configure out the lapack_lite stuff and switch
> to using the fortran versions. That has not been convered in anything
> I've personally read.
Because it's machine dependent. Here are the instructions for HP/UX:
1) Get the LAPACK/BLAS source code and compile it as position-independent
code (note: the precompiled LAPACK on netlib is not position-independent).
2) Change the Setup line for lapack_lite to
lapack_lite -DNO_APPEND_FORTRAN -I../NumPy/Include -I../NumPy/NumPyLib ../NumPy/lapack_litemodule.c -L/usr/local/lib -llapack -lblas -lcl -lm
Of course you have to change the -L option to point to your copy of
LAPACK and BLAS.
It would be nice if others could contribute appropriate instructions
for their machines for a collection of "optimization hints" for NumPy.
Konrad.
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From valb@vtek.com Tue Feb 25 14:41:21 1997
From: valb@vtek.com (Val Bykoski)
Date: Tue, 25 Feb 1997 09:41:21 -0500
Subject: [PYTHON MATRIX-SIG] switch to Explorer...
Message-ID: <199702251436.JAA20738@lucius.ultra.net>
Hi, folks!
I had all the messy problems with the download - dozens of
unsuccessful attempts. When I switched from Netscape to
Explorer 3.0, everything was fine and smooth including
installation and testing.
--Val
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Tue Feb 25 17:54:58 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Tue, 25 Feb 1997 12:54:58 -0500
Subject: [PYTHON MATRIX-SIG] NumPy test failing
Message-ID: <01BC231B.1BAB6DC0@misha.lcs.mit.edu>
Brien Barton writes:
> I installed the new NumPy 1.0b1 on an SGI system and tried to run the
> test_all.py script. It gets an error on the test that tries to verify
that,
> if a is an array, then array(a) returns a copy of a. The lines from
> test_items.py are:
>
> b = array(a)
> b[0] = 66
> a[0]
> >0
>
> This test is apparently trying to verify that after the line "b =
array(a)",
> b and a are two different objects, but on my system they are the same.
> Anyone else seeing this?
This is a bug related to the new support I added to the array function for
keyword arguments. If you use b = array(a, copy=1) explicitly this will
work. Otherwise, I've fixed this and it shouldn't be a problem in the next
beta release.
-Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Tue Feb 25 17:57:59 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Tue, 25 Feb 1997 12:57:59 -0500
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 install problem
Message-ID: <01BC231B.87CE3AA0@misha.lcs.mit.edu>
Andrew writes:
> Jim Hugunin wrote:
> >
> > As mentioned before, the file should have a size of 408,830 bytes when
> > downloaded to your system. If this is not the case, then there's
something
> > wrong with your transfer.
>
> God only knows what is going on here. I tried again, and the download
> put the file in NumPy-10-b1.txt. It wsa the right size, though, and
> after renaming, it ran, claiming to have installed NumPy. PythonWin
> doesn't seem to be all that numeric yet, since
>
> resize(0, [2, 2])
>
> results in a NameError.
>
> Maybe after rebooting this will be better.
I hate to ask the obvious question, but have you typed "from Numeric import
*"? The Numeric extensions are an extension like any other and need to be
imported into Python before they can be used.
If the problem is more complicated than this, then I'm at a loss. More
information would be helpful.
-Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Tue Feb 25 18:09:06 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Tue, 25 Feb 1997 13:09:06 -0500
Subject: [PYTHON MATRIX-SIG] inconsistency in array()?
Message-ID: <01BC231D.14724C70@misha.lcs.mit.edu>
Janko Hauser writes:
> Hello, I have tried the new NumPy-beta. I first had some problems with
> the path, I mixed up the old and the new distribution. I hope that the
> next points are unaffected by this :-)
I'm afraid that these problems sound like they are in fact related to
mixing up the old and new distributions.
> I think the asarray() function is broken, and with this the shape()
> function too. In Numeric.py (+244) array() is called with a keyword,
> wich it doesn't understand. I had a look in the old and new
> multiarray-code and the former method asarray has changed to a Numeric
> function. Is this right? And why, may I ask?
asarray has always been a function in Numeric.py. Chris Chase made some
patches available to the matrix-sig that moved it into multiarray.c to fix
some problems with user defined array classes. I adopted his patches, but
instead of adding a new C function for asarray, I modified the existing
array function to give the supported needed (without changing its previous
behavior).
In modifying the array function, I added support for keyword arguments to
it (using hooks Guido added to Python1.4). This is the first C function in
Numeric to support keyword arguments, but if all goes well it will make
things like "add.reduce(a, axis=12)" possible in future versions.
If your array function doesn't understand keyword arguments, then you are
working with an old version of the multiarray module.
> UserArray is broken but I append a version wich hopefully is ok. (It
> is broken, because it is not possible to use less() or something like
> that) There is one remaining problem, because I can't make a
> reference to a slice of the UserArray-object, I only get copys, wich
> is suboptimal. Any suggestions? :-)
I hadn't checked out UserArray closely enough (I don't use it personally).
Yours looks like a good improvement and if I don't hear any complaints
about it on the list, I'll adopt it into the next release.
-Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Tue Feb 25 19:30:44 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Tue, 25 Feb 1997 14:30:44 -0500
Subject: [PYTHON MATRIX-SIG] inconsistency in array()?
Message-ID: <01BC2328.7C51C950@misha.lcs.mit.edu>
Tom Short writes:
> >I hadn't checked out UserArray closely enough (I don't use it
personally).
> > Yours looks like a good improvement and if I don't hear any complaints
> >about it on the list, I'll adopt it into the next release.
>
> What is the purpose of UserArray? What does it allow that an Array
doesn't?
UserArray is a python class that behaves as much as possible like the lower
level array objects that are in C. The main advantage to UserArray is that
it can be subclassed to provide particular behaviors, while still behaving
as a normal array wherever possible.
The simplest example of the use of UserArray is the Matrix class that
behaves like the built-in array objects in most ways, but that reimplements
the multiply operation to do "real" matrix multiplies rather than the
normal element-wise multiplications.
I hope this answers your question - Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From amullhau@ix.netcom.com Tue Feb 25 23:38:42 1997
From: amullhau@ix.netcom.com (Andrew P. Mullhaupt)
Date: Tue, 25 Feb 1997 18:38:42 -0500
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 install problem
References: <01BC231B.8847D770@misha.lcs.mit.edu>
Message-ID: <33137802.6815@ix.netcom.com>
Jim Hugunin wrote:
>
> Andrew writes:
>
> > resize(0, [2, 2])
> >
> > results in a NameError.
> >
> > Maybe after rebooting this will be better.
>
> I hate to ask the obvious question, but have you typed "from Numeric import
> *"? The Numeric extensions are an extension like any other and need to be
> imported into Python before they can be used.
>
> If the problem is more complicated than this, then I'm at a loss. More
> information would be helpful.
Yes, I did type "from Numeric import *"
_After rebooting_, we no longer get the NameError, and what happens is
that the correct object is created, although arrays print out as
array([[1, 2, 3, 4], [5, 6, 7, 8]], 'l')
as opposed to
1 2 3 4
5 6 7 8
This problem does not occur with the version I built on my Ultrasparc
(Solaris 2.5.1).
I don't yet have a lot of code which tests NumPy so I don't really know
what works and what doesn't yet; but I don't notice any problems on the
Unix version.
Later,
Andrew Mullhaupt
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From furnish@laura.llnl.gov Wed Feb 26 16:12:55 1997
From: furnish@laura.llnl.gov (Geoffrey Furnish)
Date: Wed, 26 Feb 1997 08:12:55 -0800
Subject: [PYTHON MATRIX-SIG] HP problem
In-Reply-To: <199702250924.KAA24804@lmspc2.ibs.fr>
References: <199702242026.PAA10383@python.org>
<199702250924.KAA24804@lmspc2.ibs.fr>
Message-ID: <199702261616.LAA23323@python.org>
Konrad Hinsen writes:
> > Using gcc is not really fine except under certain conditions. If you
> > try to compile with -g, you can't make a shared module b/c of problems
> > with the size of some symbol table (and using -fPIC doesn't solve the
> > problem).
>
> True, but I see no reason to compile the LAPACK code with -g. I use
> gcc only for the two modules that cc can't handle, so that shouldn't
> cause too much trouble.
The way it happened to me seems innocent enough: I tried to build a
debuggable Python, so everything got compiled with -g, and the NumPy
stuff wouldn't compile/link. Python's build procedure is not very
well configured for the purpose of selectively compiling different
parts with different options, and linking them all together. This
seems understandable to me, since that would be a monster autoconf
job. Instead, it would seem much more reasonable if the entire code
base could be compiled -g. NumPy is the only portion of Python I am
aware of which makes this problematic. In any event, I am not
currently trying to build a debuggable Python, so for me, this is
moot at the present time.
> > Exactly /how/ does one configure out the lapack_lite stuff and switch
> > to using the fortran versions. That has not been convered in anything
> > I've personally read.
>
> Because it's machine dependent. Here are the instructions for HP/UX:
> [...]
Thanks.
--
Geoffrey Furnish email: furnish@llnl.gov
LLNL X/ICF phone: 510-424-4227 fax: 510-423-0925
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Wed Feb 26 17:02:48 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Wed, 26 Feb 1997 12:02:48 -0500
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 (printing arrays)
Message-ID: <01BC23DC.FCB0BE30@misha.lcs.mit.edu>
Andrew writes:
> Yes, I did type "from Numeric import *"
Sorry for the obvious question then; but sometimes you need to ask. I'm
also really surprised that rebooting was required after the installation.
Restarting PythonWin is sometimes needed, but I've never had to reboot
(and I install this stuff rather frequently ;-)
> _After rebooting_, we no longer get the NameError, and what happens is
> that the correct object is created, although arrays print out as
>
> array([[1, 2, 3, 4], [5, 6, 7, 8]], 'l')
>
> as opposed to
>
> 1 2 3 4
> 5 6 7 8
>
> This problem does not occur with the version I built on my Ultrasparc
> (Solaris 2.5.1).
Every array object (every python object for that matter) has two string
representations. One is considered the representation of the object
"repr(a)". This is expected to be a string that when passed in to python
will return create the object. The other is the printable string "str(a)"
for the object. This is expected to be a nice form to print out to the
user. ArrayObjects define these two representations properly (after a fair
bit of discussion to educate me on what properly is).
Your problem is that PythonWin has been written so that when you just type
"a" and return, you get "repr(a)" rather than "str(a)". This doesn't
really bother me much personally, but if you want to fix it you should
either look at the PythonWin source code, or send mail to the pythonwin-sig
asking for help.
-Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From managan@llnl.gov Wed Feb 26 17:49:39 1997
From: managan@llnl.gov (Rob Managan)
Date: Wed, 26 Feb 1997 09:49:39 -0800
Subject: [PYTHON MATRIX-SIG] Installing NumPy
Message-ID: <v03007800af3a27035096@[128.115.198.165]>
I found that when I installed NumPy 1.0b1 on an SGI that python could not
find the module until I set the environment variable to
setenv PYTHONPATH ..../Python-1.4/NumPy/NumPy
I used the instructions to build NumPy in the Python source tree. However,
the instructions only mention setting an environment variable when you
build it as a clean extension.
Is this case unusual or do the instructions need updating?
Or more to the point; what should be in the PYTHONPATH environment variable
and what is it used for?
*-*-*-*-*-*-*-*-*-*-**-*-*-*-*-*-*-*-*-*-*-
Rob Managan mailto://managan@llnl.gov
LLNL ph: 510-423-0903
P.O. Box 808, L-183 FAX: 510-423-5804
Livermore, CA 94551-0808
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hinsen@ibs.ibs.fr Wed Feb 26 18:50:27 1997
From: hinsen@ibs.ibs.fr (Konrad Hinsen)
Date: Wed, 26 Feb 1997 19:50:27 +0100
Subject: [PYTHON MATRIX-SIG] Installing NumPy
In-Reply-To: <v03007800af3a27035096@[128.115.198.165]> (message from Rob
Managan on Wed, 26 Feb 1997 09:49:39 -0800)
Message-ID: <199702261850.TAA31112@lmspc2.ibs.fr>
> I used the instructions to build NumPy in the Python source tree. However,
> the instructions only mention setting an environment variable when you
> build it as a clean extension.
>
> Is this case unusual or do the instructions need updating?
Yes. See my updated instructions for details ;-)
> Or more to the point; what should be in the PYTHONPATH environment variable
> and what is it used for?
Python looks for modules in all directories that are listed in sys.path.
This variable is initialized to the value of PYTHONPATH followed by
the list of directories precompiled into the interpreter. The idea
is that the latter should contain everything needed for the standard
installation, and PYTHONPATH the user-specific additions. But of
course the borderline is not entirely clear.
In fact, I use the "source tree" installation method mostly because I want
the NumPy path to be built-in, since otherwise users without PYTHONPATH
couldn't use it.
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@ibs.ibs.fr
Laboratoire de Dynamique Moleculaire | Tel.: +33-4.76.88.99.28
Institut de Biologie Structurale | Fax: +33-4.76.88.54.94
41, av. des Martyrs | Deutsch/Esperanto/English/
38027 Grenoble Cedex 1, France | Nederlands/Francais
-------------------------------------------------------------------------------
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From pas@xis.xerox.com Wed Feb 26 19:33:56 1997
From: pas@xis.xerox.com (Perry A. Stoll)
Date: Wed, 26 Feb 1997 11:33:56 PST
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 (printing arrays)
In-Reply-To: <01BC23DC.FCB0BE30@misha.lcs.mit.edu>
Message-ID: <199702261934.OAA28684@fantod.xis.xerox.com>
On 26 Feb, Jim Hugunin wrote:
> [accurate description of repr/str snipped]
> Your problem is that PythonWin has been written so that when you just type
> "a" and return, you get "repr(a)" rather than "str(a)". This doesn't
> really bother me much personally, but if you want to fix it you should
> either look at the PythonWin source code, or send mail to the pythonwin-sig
> asking for help.
I don't think it's only PythonWin! On SunOS I get:
jade{pas}<exp1/train> python
Python 1.4 (Jan 22 1997) [GCC 2.7.0]
Copyright 1991-1995 Stichting Mathematisch Centrum, Amsterdam
>>> class foo:
... def __repr__(self): return 'repr called'
... def __str__(self): return 'str called'
...
>>> f = foo()
>>> f
repr called
>>> print f
str called
>>>
-Perry
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From amullhau@ix.netcom.com Thu Feb 27 04:44:31 1997
From: amullhau@ix.netcom.com (Andrew P. Mullhaupt)
Date: Wed, 26 Feb 1997 23:44:31 -0500
Subject: [PYTHON MATRIX-SIG] Win95 NumPy 1.0b1 (printing arrays)
References: <199702261934.OAA28684@fantod.xis.xerox.com>
Message-ID: <3315112F.1F26@ix.netcom.com>
Perry A. Stoll wrote:
>
> On 26 Feb, Jim Hugunin wrote:
>
> > [accurate description of repr/str snipped]
>
> I don't think it's only PythonWin! On SunOS I get:
Well, you can't sell me much enthusiasm for choosing repr/str when
neither one is worth a damn for functions.
If we have
def f() : return 1
then it seems that repr(f) ought to be
'def f() : return 1'
and not
'<function f at ....>'
which, no matter how many times you type it in is not going to get
you your f back.
Later,
Andrew Mullhaupt
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Thu Feb 27 20:30:52 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Thu, 27 Feb 1997 15:30:52 -0500
Subject: [PYTHON MATRIX-SIG] The difference between repr and str, and printing arrays
Message-ID: <9702272031.AA01122@goldilocks>
I think I just fixed a bug in the way that PythonWin prints out objects in
its interpreter. The cross-posting is to make sure I not instead confused
about the "proper" way for things to work. Here's the issue (this uses the
Numeric extensions to python, but should be easy enough to follow for
anyone interested in repr's and str's). Under PythonWin's interpreter I do
the following:
>>> from Numeric import *
>>> a = array([1,2,3])
>>> a
array([1, 2, 3], 'l')
>>> repr(a)
"array([1, 2, 3], 'l')"
>>> str(a)
'1 2 3'
Notice that when I just type "a", I'm getting its "repr", rather than the
printable string version (which is what I want).
Under most other interpreters (DOS python for example) I get:
>>> from Numeric import *
>>> a = array([1,2,3])
>>> a
1 2 3
>>> repr(a)
"array([1, 2, 3], 'l')"
>>> str(a)
'1 2 3'
This is the behavior I'd hoped for (and designed the repr/str conventions
of NumPy around).
The discrepancy is based on the fact that PythonWin prints out the
repr(eval(command)) instead of either the str(eval(command)) or just
eval(command) when something is typed to it's interactive prompt.
I can fix PythonWin by changing line 317 of interact.py to read "print ret"
or "print str(ret)" instead of "print repr(ret)".
When I fix it, I get the following:
>>> from Numeric import *
>>> a = arange(10)
>>> a
0 1 2 3 4 5 6 7 8 9
>>> repr(a)
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], 'l')
>>> str(a)
0 1 2 3 4 5 6 7 8 9
Notice that the arrays are now printing as I would wish, but that the
quotes have been removed from around the strings in the results. I happen
to like this appearence in general, but I wanted to understand two things:
1) Why does there not seem to be a simple way to get both strings and
arrays to print out the way they do in most command-line interpreters?
2) Should this fix be applied to PythonWin? Does it do the right thing,
and would it break anything?
Thanks for any comments - Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Fri Feb 28 19:38:06 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Fri, 28 Feb 1997 14:38:06 -0500
Subject: [PYTHON MATRIX-SIG] Printing arrays and repr vs. str
Message-ID: <01BC2585.032B78F0@misha.lcs.mit.edu>
PythonWin has just highlighted a critical flaw in the design of array
printing for NumPy.
To explain: For C objects, there are three methods related to the string
representation of an array. One returns the str(a), one returns the
repr(a), and the third one is used to print out the array directly.
Under the current design, str(a) and repr(a) both behave "properly" for
python's design. Unfortunately, the print method uses the str(a)
representation, which according to Guido is not the correct way to display
python objects (and if anybody should know it's Guido ;-).
The interesting fact is that this "bug" in the print method has led to the
behavior that almost all users of NumPy seem to want in their interactive
work with arrays.
I am at a total loss as to how to fix this. The right thing to do is to
change the behavior of the print method to use the repr version of an
array. This will however make it much less enjoyable to work with arrays
in an interpreter. Not fixing the bug in the print method though will make
NumPy inconsistent with the rest of python objects, and lead to
inconsistent behavior as observed in PythonWin recently.
Does anybody have any good suggestions on this one?
-Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From guido@CNRI.Reston.Va.US Fri Feb 28 19:57:32 1997
From: guido@CNRI.Reston.Va.US (Guido van Rossum)
Date: Fri, 28 Feb 1997 14:57:32 -0500
Subject: [PYTHON MATRIX-SIG] Re: Printing arrays and repr vs. str
In-Reply-To: Your message of "Fri, 28 Feb 1997 14:38:06 EST."
<01BC2585.032B78F0@misha.lcs.mit.edu>
References: <01BC2585.032B78F0@misha.lcs.mit.edu>
Message-ID: <199702281957.OAA01885@monty>
> Under the current design, str(a) and repr(a) both behave "properly" for
> python's design. Unfortunately, the print method uses the str(a)
> representation, which according to Guido is not the correct way to display
> python objects (and if anybody should know it's Guido ;-).
I'm sorry that it's so complicated, but this is still not quite
right. (I had to browse the sources myself, ehm.)
When an object is being "printed" to a "real file", its print method
is invoked with a flag. The flag can be 0 or Py_PRINT_RAW. When it
is 0, you are supposed to emulate repr(); when it is Py_PRINT_RAW, you
are supposed to emulate str(). For some details, see print_object()
in Objects/object.c. (Actually, it's a word of flag bits.)
When an object is being "printed" to something that only emulates a
file, Python calls repr() or str() directly. This also happens when
the object does not have a print method.
The print statement uses Py_PRINT_RAW. Expressions entered at the >>>
prompt use a 0 flag.
> Does anybody have any good suggestions on this one?
Have a user preference flag that determines whether repr() should
behave like str()? Set it to on by default when Python is used
interactively.
--Guido van Rossum (home page: http://www.python.org/~guido/)
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________
From hugunin@mit.edu Thu Feb 27 20:30:52 1997
From: hugunin@mit.edu (Jim Hugunin)
Date: Thu, 27 Feb 1997 15:30:52 -0500
Subject: [PYTHON MATRIX-SIG] The difference between repr and str, and printing arrays
Message-ID: <9702272031.AA01122@goldilocks>
I think I just fixed a bug in the way that PythonWin prints out objects in
its interpreter. The cross-posting is to make sure I not instead confused
about the "proper" way for things to work. Here's the issue (this uses the
Numeric extensions to python, but should be easy enough to follow for
anyone interested in repr's and str's). Under PythonWin's interpreter I do
the following:
>>> from Numeric import *
>>> a = array([1,2,3])
>>> a
array([1, 2, 3], 'l')
>>> repr(a)
"array([1, 2, 3], 'l')"
>>> str(a)
'1 2 3'
Notice that when I just type "a", I'm getting its "repr", rather than the
printable string version (which is what I want).
Under most other interpreters (DOS python for example) I get:
>>> from Numeric import *
>>> a = array([1,2,3])
>>> a
1 2 3
>>> repr(a)
"array([1, 2, 3], 'l')"
>>> str(a)
'1 2 3'
This is the behavior I'd hoped for (and designed the repr/str conventions
of NumPy around).
The discrepancy is based on the fact that PythonWin prints out the
repr(eval(command)) instead of either the str(eval(command)) or just
eval(command) when something is typed to it's interactive prompt.
I can fix PythonWin by changing line 317 of interact.py to read "print ret"
or "print str(ret)" instead of "print repr(ret)".
When I fix it, I get the following:
>>> from Numeric import *
>>> a = arange(10)
>>> a
0 1 2 3 4 5 6 7 8 9
>>> repr(a)
array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], 'l')
>>> str(a)
0 1 2 3 4 5 6 7 8 9
Notice that the arrays are now printing as I would wish, but that the
quotes have been removed from around the strings in the results. I happen
to like this appearence in general, but I wanted to understand two things:
1) Why does there not seem to be a simple way to get both strings and
arrays to print out the way they do in most command-line interpreters?
2) Should this fix be applied to PythonWin? Does it do the right thing,
and would it break anything?
Thanks for any comments - Jim
_______________
MATRIX-SIG - SIG on Matrix Math for Python
send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________