Mailman 3 September 2007 - NumPy-Discussion

latest svn version fails on Solaris
by Christopher Hanley 20 Sep '07

20 Sep '07

Hi, The latest version of numpy has a unit test failure on big endian machines. ====================================================================== FAIL: test_record_array (numpy.core.tests.test_multiarray.test_putmask) ---------------------------------------------------------------------- Traceback (most recent call last): File "/data/basil5/site-packages/lib/python/numpy/core/tests/test_multiarray.py", line 450, in test_record_array assert_array_equal(rec['x'],[10,5]) File "/data/basil5/site-packages/lib/python/numpy/testing/utils.py", line 223, in assert_array_equal verbose=verbose, header='Arrays are not equal') File "/data/basil5/site-packages/lib/python/numpy/testing/utils.py", line 215, in assert_array_compare assert cond, msg AssertionError: Arrays are not equal (mismatch 50.0%) x: array([ 4.58492919e-320, 5.00000000e+000]) y: array([10, 5]) ---------------------------------------------------------------------- Ran 670 tests in 47.182s Chris

3 2

Find first occurrence?
by Zachary Pincus 20 Sep '07

20 Sep '07

Hello all, On several occasions, I've had the need to find only the first occurrence of a value in an unsorted numpy array. I usually use numpy.where(arr==val)[0] or similar, and don't worry about the fact that I'm iterating across the entire array. However, sometimes the arrays are pretty big and the find operation is in an inner loop, so I was wondering if there's already a C extension function somewhere in numpy or scipy that does a fast find first operation, or anything similar. (Easy enough to write my own, and maybe given the issues inherent in comparing float equality, etc., it doesn't belong in the core numpy anyway...) Thanks, Zach Pincus Program in Biomedical Informatics and Department of Biochemistry

1 0

confusion about min/max
by mark 19 Sep '07

19 Sep '07

Hello - When I am doing from numpy import * It does not import the min() function, but when I do from numpy import min it does import the min() function Does that make sense? I know, I should probably use a.min() rather than min(a), but why does min() not get imported on an import * ? Thanks, Mark

4 4

ANN: Trilinos 8.0, including PyTrilinos 4.0
by Bill Spotz 19 Sep '07

19 Sep '07

Version 8.0 of Trilinos has been released: http://trilinos.sandia.gov Trilinos is a collection of scientific, object-oriented solver packages. These packages cover linear algebra services, preconditioners, linear solvers, nonlinear solvers, eigensolvers, and a wide range of related utilities. Trilinos supports serial and parallel architectures, as well as dense or sparse problem formulations. Included in Trilinos release 8.0 is PyTrilinos version 4.0, http://trilinos.sandia.gov/packages/pytrilinos a set of python interfaces to selected Trilinos packages. New in version 4.0 of PyTrilinos is an interface to Anasazi, the eigensolver package, and the re-enabling of NOX, the nonlinear solver package. The primary Trilinos linear algebra services package is Epetra, which provides Vector and MultiVector classes, as well as hierarchies of operator, communicator and domain decomposition classes. The PyTrilinos interface to Epetra has been designed with a high degree of compatibility with numpy, with the hope of complementing the SciPy development efforts. ** Bill Spotz ** ** Sandia National Laboratories Voice: (505)845-0170 ** ** P.O. Box 5800 Fax: (505)284-5451 ** ** Albuquerque, NM 87185-0370 Email: wfspotz(a)sandia.gov **

1 0

FPE on tensordot
by Alexandre Fayolle 18 Sep '07

18 Sep '07

Hi, A user of some code I've written is experiencing some strange behaviour with numpy.tensordot. I have unfortunately no access to his computer and cannot reproduce the crash on my machine. The short way of reproducing this is: import numpy a=numpy.array([0.5,0.5]) b=numpy.array([[0.,1.],[2.,3.]]) numpy.tensordot(a,b,axes=(0,0)) This works and returns array([ 1., 2.]) import numpy a=numpy.array([0.4,0.5]) b=numpy.array([[0.,1.],[2.,3.]]) numpy.tensordot(a,b,axes=(0,0)) This crashes with an FPE. He is running Linux (Debian sarge based) with python 2.3.5. Numpy 1.0.3.1 was compiled manually by the admins of the lab (and I don't know which options they used, probably the default). The machine is a 32bit Intel. I'm puzzled, and welcome any insight on this. Regards -- Alexandre Fayolle LOGILAB, Paris (France)

1 0

Segfault on wrong use of indexing
by Arnar Flatberg 17 Sep '07

17 Sep '07

Hi list A pretty common use (for me) is to create arrays at the beginning of my code and then fill in parts as I go along. Today, my code hit a segmentation fault. The error was that an index-vector ,that usually is a list with several members, now contained only one member and I tried to insert a 1-dim vector. Here is a short example of what went wrong: Python 2.5.1 (r251:54863, May 2 2007, 16:56:35) [GCC 4.1.2 (Ubuntu 4.1.2-0ubuntu4)] on linux2 Type "help", "copyright", "credits" or "license" for more information. >>> import numpy as n >>> n.__version__ '1.0.1' >>> a = n.random.rand(4,2) >>> a array([[ 0.79561255, 0.4932522 ], [ 0.84008255, 0.88258233], [ 0.35143166, 0.82214487], [ 0.12490114, 0.15799074]]) Let us insert [2,2] into the first row. Correct use: >>> a[0,:] = [2,2] >>> a array([[ 2. , 2. ], [ 0.84008255, 0.88258233], [ 0.35143166, 0.82214487], [ 0.12490114, 0.15799074]]) or: a[[0],:] = [[2,2]] However, when I tried to to insert a 1-dim array with a 'two-dim' index things went wrong: >>> a[[0],:] = [2,2] Segmentation fault Sometimes, (with some lucky pointers, I guess) the latter code will run, with incorrect values in the array. With incorrect, I mean compared to the correct use of indices :-) My fix was to call "atleast_2d" on the parts I wanted to insert into the array. Am I just a horrible user of indices, or is this a bug? The situation were incorrect use of indices just passes silently is what I find a little disturbing. Thanks, Arnar

3 3

Re: [Numpy-discussion] Find contiguous sequences of booleans in arrays
by Charles R Harris 17 Sep '07

17 Sep '07

Hi Kurdt, On 9/17/07, Kurdt Bane <kurdt.bane(a)gmail.com> wrote: > > Hi to all, > > I've got an 1-D array of bools and I'd like to find the length of the > first contiguous sequence of True values, starting from position [0] of the > array. > (That's equivalent to find the position of the first occurrence of False > in the array). > The problem is trivial, but I was wondering: what's the best (fastest, > cleanest, most pythonesque) way to do it in numpy? And what if I want to get > a list of all the contiguous sequences of True values, above a given > threshold? You can find the start of all runs after the first by In [1]: a = array([1,1,1,0,1,1,0,0,1], dtype=bool) In [2]: s = arange(1,len(a))[a[0:-1] ^ a[1:]] In [3]: s Out[3]: array([3, 4, 6, 8]) i.e. The first run is a[0:3], the second a[3:4], etc., and the runs alternate between true and false. Chuck

1 0

‎ Find contiguous sequences of booleans in arrays
by Kurdt Bane 17 Sep '07

17 Sep '07

Hi to all, I've got an 1-D array of bools and I'd like to find the length of the first contiguous sequence of True values, starting from position [0] of the array. (That's equivalent to find the position of the first occurrence of False in the array). The problem is trivial, but I was wondering: what's the best (fastest, cleanest, most pythonesque) way to do it in numpy? And what if I want to get a list of all the contiguous sequences of True values, above a given threshold? Thanks in advance for your advices, Chris.

3 2

Is this a bug?
by Charles R Harris 16 Sep '07

16 Sep '07

I note a small inconsistency in the use of the out keyword in some functions: >>> a=array(0) >>> sometrue([1],out=a).shape () >>> a=array([0]) >>> sometrue([1],out=a).shape (1,) >>> a=array([[0]]) >>> sometrue([1],out=a).shape (1, 1) >>> a=array([[0,0]]) >>> sometrue(eye(2),axis=1,out=a).shape (1, 2) It seems to me that all but the first case should raise an error, as the shape of the output array is not the same as the expected output. I know this looks picky, but 0d arrays can't be indexed, whereas 1d and 2d arrays can, so they aren't quite compatible. Besides, the current behavior is difficult to describe for the documentation. If the current behavior is the rule, what is that rule? Chuck

2 1

arange and floating point arguments
by Ed Schofield 16 Sep '07

16 Sep '07

Hi everyone, This was reported yesterday as a bug in Debian's numpy package: >>> len(numpy.arange(0, 0.6, 0.1)) == len(numpy.arange(0, 0.4+0.2, 0.1)) False The cause is this: >>> ceil((0.4+0.2)/0.1) 7.0 >>> ceil(0.6/0.1) 6.0 which holds for both numpy's and the standard library's ceil(). Using arange in this way is a fundamentally unreliable thing to do, but is there anything we want to do about this? Should numpy emit a warning when using arange with floating point values when (stop-start)/step is close to an integer? -- Ed

12 22