Mailman 3 April 2006 - NumPy-Discussion

sob story
by Bert Morrow 27 Apr '06

27 Apr '06

1 0

Warnings in NumPy SVN
by Travis Oliphant 27 Apr '06

27 Apr '06

I want to apologize for the relative instability of the SVN tree in the past couple of days. Getting the scalarmath layout working took more C-API changes than I had anticipated. The SVN version of NumPy now builds scalarmath by default. The basic layout of the module is complete. However, there are many basic functions that are missing. As a result, during compile you will get many warnings about undefined functions. If an attempt were made to load the module it would cause an error as well due to undefined symbols. These undefined symbols are all the basic operations on fundamental c data-types that either need a function defined or a #define statement made. The names have this form: @name@_ctype_@oper@ where @name@ is one of the 16 Number-like types and @oper@ is one of the operations needing to be supported. The function (or macro) needs to implement the operation on the basic data-type and if necessary set an error-flag in the floating-point registers. If anybody has time to help implement these basic operations, it would be greatly appreciated. -Travis

5 6

Re: [Numpy-discussion] Broadcasting rules (Ticket 76).
by tim.hochberg＠cox.net 27 Apr '06

27 Apr '06

---- Travis Oliphant <oliphant.travis(a)ieee.org> wrote: > Sasha wrote: > > In this category, I would suggest to allow broadcasting to any > > multiple of the dimension even if the dimension is not 1. I don't see > > what makes 1 so special. > > > What's so special about 1 is that the code for it is relatively > straightforward and already implemented using strides. Altering the > code to allow any multiple of the dimension would be harder and slower. It also does the right thing most of the time and is easy to understand. It's my expectation that oppening up broadcasting will be more effective in masking errors than in enabling useful new behaviour. I think that's my ticket being discussed here. If so, it was motivated by a case that stopped working because the looser broadcasting behaviour was preventing some other broadcasting from taking place. I'm not home right now, so I can't provide details; I'll do that on Thursday. Just keep in mind that it's much easier to keep the broadcasting rules restrictive for now and loosen them up later than to try to tighten them up later if loosening them up turns out to not be a good idea. -tim

8 16

matrix.std() returns array
by Keith Goodman 27 Apr '06

27 Apr '06

I noticed that the mean of a matrix is a matrix but the standard deviation of a matrix is an array. Is that the expected behavior? I'm also getting the wrong values (0 and nan) for the standard deviation. Did I mess something up? I'm trying to learn scipy (and python) by porting a small Octave program. I installed numpy from svn (today) on a Debian box. And numpy.test() says OK. Here's an example: >> numpy.__version__ '0.9.7.2416' >> x = asmatrix(random.uniform(0,1,(3,3))) >> x matrix([[ 0.56771284, 0.57053769, 0.57505946], [ 0.10479534, 0.81692248, 0.91829316], [ 0.48627829, 0.59255983, 0.32628573]]) >> x.mean(0) matrix([[ 0.38626216, 0.66000667, 0.60654612]]) >> x.std(0) array([ nan, 0. , 0. ])

5 5

equality with masked object
by Lionel Roubeyrie 27 Apr '06

27 Apr '06

Hi all, I have a problem with masked_object (and masked_values to) like in this sort example : ########################################### lionel[Données]8>test=array([1,2,3,inf,5]) lionel[Données]9>test = ma.masked_object(test, inf) lionel[Données]10>print test[3], type(test[3]) -- <class 'numpy.core.ma.MaskedArray'> lionel[Données]11>print test.max(), type(test.max()) 5.0 <type 'float64scalar'> lionel[Données]12>test[3] == test.max() Sortie[12]: array(data = [True], mask = True, fill_value=?) ########################################### Why 5.0 == -- return True? A float is it the same as a masked object? thanks -- Lionel Roubeyrie - lroubeyrie(a)limair.asso.fr LIMAIR http://www.limair.asso.fr

2 3

array.min() vs. min(array)
by Ryan Krauss 27 Apr '06

27 Apr '06

I was spending some time trying to track down how to speed up an algorithm that gets called a bunch of times during an optimization. I was startled when I finally figured out that most of the time was wasted by using the built-in pyhton min function. It turns out that in my case, using array.min() (i.e. the method of the Numpy array) is 300-500 times faster than the built-in python min function (i.e. min(array)). So, thank you Travis and everyone who has put so much time into thinking through Numpy and making it fast (as well as making sure it is correct). And to the rest of us, use the Numpy array methods whenever you can. Thanks, Ryan

6 9

vectorize problem
by Steffen Loeck 26 Apr '06

26 Apr '06

Hello all, I have a problem using scalar variables in a vectorized function: from numpy import vectorize def f(x): if x>0: return 1 else: return 0 F = vectorize(f) F(1) gives the error message: --------------------------------------------------------------------------- exceptions.AttributeError Traceback (most recent call last) .../function_base.py in __call__(self, *args) 619 620 if self.nout == 1: --> 621 return self.ufunc(*args).astype(self.otypes[0]) 622 else: 623 return tuple([x.astype(c) for x, c in zip(self.ufunc(*args), self.otypes)]) AttributeError: 'int' object has no attribute 'astype' Is there any way to get vectorized functions working with scalars again? Regards Steffen

3 3

concatenate, doc-string
by Arnd Baecker 26 Apr '06

26 Apr '06

Hi, the doc-string of concatentate is pretty short: numpy.concatenate? Docstring: concatenate((a1,a2,...),axis=None). Would the following be better: """ concatenate((a1, a2,...), axis=None) joins the tuple `(a1, a2, ...)` of sequences (or arrays) into a single numpy array. Example:: print concatenate( ([0,1,2], [5,6,7])) """ ((The ``(or arrays)`` could be omitted if sequences include array by default, though it might not be obvious to beginners ...)) I was also tempted to suggest a dtype argument, concatenate( ([0,1,2], [5,6,7]), dtype=numpy.Float) but I am not sure if that would be a good idea ... Best, Arnd

4 5

SWIG for 3D array
by Gennan Chen 26 Apr '06

26 Apr '06

Hi! I will like to use SWIG to wrap my code. However, it seems the current numpy.i only can map 1 and 2D array, but not 3D. Is it correct? Or I miss something here. I don't mind spend some time to do it like scipy.ndimage if numpy.i did not support ND arrary. But I am new to write extension to Python. And I really have hard time to understand how to deal with reference counting issues. Anyone know where I can know a good reference for that? Or a simple example in numpy will be appreciated.... Gen-Nan Chen, PhD Chief Scientist Research and Development Group CorTechs Labs Inc (www.cortechs.net) 1020 Prospect St., #304, La Jolla, CA, 92037 Tel: 1-858-459-9700 ext 16 Fax: 1-858-459-9705 Email: gnchen(a)cortechs.net

2 1

crush!!!!
by Hugo Siles 26 Apr '06

26 Apr '06

HI, I have a problem when I run the following options in python: >>>from Numeric import * >>>from Linear algebra I define a matrix 'a' which prints correctly, calculates its inverse, determinat and so for but when I try to calculate the eigenvalues, such as >>> c = eigenvalues(a) the system just crushs without any message I made this test because in some other programs with source code happens the same thing. I hope some body can help, thanks Hugo Siles

2 1