Mailman 3 August 2011 - NumPy-Discussion

Wrong treatment of byte-order.
by Nadav Horesh 31 Aug '11

31 Aug '11

Hi, This is my second post on this problem I found in numpy 1.6.1, and recently it cam up in the latest git version (2.0.0.dev-f3e70d9). The problem is numpy treats the native byte order ('<') as illegal while the wrong one ('>') as the right one. The output of the attached script (bult for python 2.6 + ) is given below (my system is a 64 bit linux on core i7. 64 bit python 2.7.2/3.2 , numpy uses ATLAS): $ python test_byte_order.py a = [[ 0.28596132 0.31658824 0.34929676] [ 0.48739246 0.68020533 0.39616588] [ 0.29310406 0.9584545 0.8120068 ]] a1 = [[ 0.28596132 0.31658824 0.34929676] [ 0.48739246 0.68020533 0.39616588] [ 0.29310406 0.9584545 0.8120068 ]] (Wrong byte order on Intel CPUs): a2 = [[ 8.97948198e-017 1.73406416e-025 -4.25909057e+014] [ 4.59443694e+090 7.91693101e-029 5.26959329e-135] [ 2.93240450e+060 -2.25898860e-051 -2.06126917e+302]] Invert a: OK Invert a2 (Wrong byte order!): OK invert a1: Traceback (most recent call last): File "test_byte_order.py", line 20, in <module> b1 = N.linalg.inv(a1) File "/usr/lib64/python2.7/site-packages/numpy/linalg/linalg.py", line 445, in inv return wrap(solve(a, identity(a.shape[0], dtype=a.dtype))) File "/usr/lib64/python2.7/site-packages/numpy/linalg/linalg.py", line 326, in solve results = lapack_routine(n_eq, n_rhs, a, n_eq, pivots, b, n_eq, 0) lapack_lite.LapackError: Parameter a has non-native byte order in lapack_lite.dgesv

1 0

nan division warnings
by mdekauwe 31 Aug '11

31 Aug '11

Hi, this is probably my lack of understanding...when i set up some masks for 2 arrays and try to divide one by the other I get a runtime warning. Seemingly this is when I am asking python to divide one nan by the other, however I thought by masking the array numpy would then know to ignore these nans? For example import numpy as np a = np.array([4.5, 6.7, 8.0, 9.0, 0.00001]) b = np.array([0.0001, 6.7, 8.0, 9.0, 0.00001]) a = np.ma.where(np.logical_or(a<0.01, b<0.01), np.nan, a) b = np.ma.where(np.logical_or(a<0.01, b<0.01), np.nan, b) a/b will produce …./numpy/ma/core.py:772: RuntimeWarning: invalid value encountered in absolute return umath.absolute(a) * self.tolerance >= umath.absolute(b) but of course give the correct result masked_array(data = [-- 1.0 1.0 1.0 --], mask = [ True False False False True], fill_value = 1e+20) But what is the correct way to do this array division such that I don't produce the warning? The only way I can see that you can do it is a bit convoluted and involves empty the array of the masked values, e.g. a = a[np.isnan(a) == False] b = b[np.isnan(b) == False] a/b thanks, Martin -- View this message in context: http://old.nabble.com/nan-division-warnings-tp32369310p32369310.html Sent from the Numpy-discussion mailing list archive at Nabble.com.

2 2

Issue with dtype and nx1 arrays
by Thomas Robitaille 31 Aug '11

31 Aug '11

Hello, Is the following behavior normal? In [1]: import numpy as np In [2]: np.dtype([('a','<f4',2)]) Out[2]: dtype([('a', '<f4', (2,))]) In [3]: np.dtype([('a','<f4',1)]) Out[3]: dtype([('a', '<f4')]) I.e. in the second case, the second dimension of the dtype (1) is being ignored? Is there a way to avoid this? Thanks, Thomas

2 1

numpy oddity
by Johann Cohen-Tanugi 30 Aug '11

30 Aug '11

I have numpy version 1.6.1 and I see the following behavior : In [380]: X Out[380]: 1.0476157527896641 In [381]: X.__class__ Out[381]: numpy.float64 In [382]: (2,3)*X Out[382]: (2, 3) In [383]: (2,3)/X Out[383]: array([ 1.90909691, 2.86364537]) In [384]: X=float(X) In [385]: (2,3)/X --------------------------------------------------------------------------- TypeError Traceback (most recent call last) /home/cohen/<ipython-input-385-cafbe080bfd5> in <module>() ----> 1 (2,3)/X TypeError: unsupported operand type(s) for /: 'tuple' and 'float' So it appears that X being a numpy float allows numpy to play some trick on the tuple so that division becomes possible, which regular built-in float does not allow arithmetics with tuples. But why is multiplication with "*" not following the same prescription? best, Johann

4 6

numpy-1.6.1.win32-py3.2.exe (md5) won't install
by Richard D. Moores 30 Aug '11

30 Aug '11

Python 3.2, 64-bit Win 7 When I try to install numpy-1.6.1.win32-py3.2.exe (md5) I get "Python version 3.2 required, which was not found in the registry". What to do? Thanks, Dick Moores

4 13

wierd numpy.void behavior
by Neal Becker 30 Aug '11

30 Aug '11

I've encountered something weird about numpy.void. arr = np.empty ((len(results),), dtype=[('deltaf', float), ('quantize', [('int', int), ('frac', int)])]) for i,r in enumerate (results): arr[i] = (r[0]['deltaf'], tuple(r[0]['quantize_mf'])) from collections import defaultdict, namedtuple experiments = defaultdict(list) testcase = namedtuple ('testcase', ['quantize']) for e in arr: experiments[testcase(e['quantize'])].append (e) Now it seems that when e['quantize'] is used as a dictionary key, equal values are not compared as equal: In [36]: experiments Out[36]: defaultdict(<type 'list'>, {testcase(quantize=(0, 0)): [(1.25, (0, 0))], testcase(quantize=(0, 0)): [(1.25, (0, 0))], testcase(quantize=(0, 0)): [(1.25, (0, 0))]}) See, there are 3 'testcases' inserted, all with keys quantize=(0,0). In [37]: e['quantize'] Out[37]: (0, 0) In [38]: type(e['quantize']) Out[38]: <type 'numpy.void'> There's something weird here. If instead I do: for e in arr: experiments[testcase(tuple(e['quantize']))].append (e) that is, convert e['quantize'] to a tuple before using it as a key, I get the expected behavior: In [40]: experiments Out[40]: defaultdict(<type 'list'>, {testcase(quantize=(0, 0)): [(1.25, (0, 0)), (1.25, (0, 0)), (1.25, (0, 0))]})

2 1

Want to find a scientific app using NumPy to process large set of data, say more than 1000000 elements in ndarray.
by Qisheng Yang 30 Aug '11

30 Aug '11

Hello, All As the subject say, I want to exercise *multiprocessing *module in NumPy in order to take advantage of multi-cores. A project which processing large set of data will be useful to compare single thread with multi-thread. I have reviewed some projects using NumPy/SciPy list on SciPy homepage. But I haven't yet found a project which using NumPy ufunc to process large set of data. Any suggestions would be greatly appreciated. Thanks much.

1 0

numpy.log does not raise exceptions
by Dominique Orban 28 Aug '11

28 Aug '11

Hi, I'm wondering why numpy.log doesn't raise a ValueError exception the way math.log does: 1< import numpy as np 2< np.log([-1]) Warning: invalid value encountered in log 2> array([ nan]) 3< import math 4< math.log(-1) --------------------------------------------------------------------------- ValueError Traceback (most recent call last) It would make it a lot easier to trap domain errors than using isnan(). Thanks, -- Dominique

3 2

Removing numscons, adding bento scripts to main branch ?
by David Cournapeau 27 Aug '11

27 Aug '11

Hi there, I am finally at a stage where bento can do most of what numscons could do. I would rather avoid having 3 different set of build scripts (distutils+bento+numscons) to maintain in the long term, so I would favor removing numscons scripts from numpy and scipy. I was thinking about keeping maybe numscons scripts for one release for both numpy/scipy, with a warning about their deprecation, and then removing them one release later. Does that sound ok with everyone ? cheers, David

6 6

the build and installation process
by srean 26 Aug '11

26 Aug '11

Hi, I would like to know a bit about how the installation process works. Could you point me to a resource. In particular I want to know how the site.cfg configuration works. Is it numpy/scipy specific or is it standard with distutils. I googled for site.cfg and distutils but did not find any authoritative document. I believe many new users trip up on the installation process, especially in trying to substitute their favourite library in place os the standard. So a canonical document explaining the process will be very helpful. http://docs.scipy.org/doc/numpy/user/install.html does cover some of the important points but its a bit sketchy, and has a "this is all that you need to know" flavor. Doesnt quite enable the reader to fix his own problems. So a resource that is somewhere in between reading up all the sources that get invoked during the installation and building, and the current install document will be very welcome. English is not my native language, but if there is anyway I can help, I would do so gladly. -- srean

2 1

NumPy-Discussion August 2011