Mailman 3 February 2009 - NumPy-Discussion

Sphinx custom extension mess, and patches
by Gael Varoquaux Feb. 16, 2009

Feb. 16, 2009

Hi all, Sorry for the multiple posting, this concerns various groups, and I'd rather the information not be lost. While working on getting our in-lab library ready to be merged with NiPy, I ran into some sort of 'sphinx extension mess' where various sphinx extension would have side effects on each other, and most important, the extensions did not work with sphinx trunk. I got the side effects to be limited by cleaning up the generated code from autosummary before each run: I added the following code in my sphinx conf.py: ################################################################################ # Hack: run the autosummary generation script import shutil if os.path.exists('generated'): shutil.rmtree('generated') os.system('%s sphinxext/autosummary_generate.py -o generated *.rst' % sys.executable) ################################################################################ I am attaching a diff of all the modifications I made to get the various extensions to work. I hope you can use it to get your various extensions working on sphinx trunk quicker. For the NiPy guys, I will be committing these changes in the fff2 tree soon, and we can go over this at the sprint. This does raise a problem: this extension code is all over the place, in various repository. Some of the code cannot live in the sphinx repo, as it introduces dependencies. However, as the extensions are not importable from Python (I can't do 'from matplotlib.sphinxext import mathmpl'), the different projects using them end up copying them in their repo, and thus there are several versions floating around not updated. Some of the extensions would do not add externa dependencies to sphinx. These should be pushed into sphinx, with tests. That way as sphinx evolves, they do not break. Gaël

5 8

numpy.linalg or scipy.linalg ?
by axel.breuer＠bnpparibas.com Feb. 16, 2009

Feb. 16, 2009

Hi, Which one is better numpy.linalg or scipy.linalg ? These modules have some common functions but with different help lines. Furthermore, numpy.scipy.linalg.svd seems to be more stable than numpy.linalg.svd (are they using different LAPACK methods ?) This message and any attachments (the "message") is intended solely for the addressees and is confidential. If you receive this message in error, please delete it and immediately notify the sender. Any use not in accord with its purpose, any dissemination or disclosure, either whole or partial, is prohibited except formal approval. The internet can not guarantee the integrity of this message. BNP PARIBAS (and its subsidiaries) shall (will) not therefore be liable for the message if modified. Do not print this message unless it is necessary, consider the environment. --------------------------------------------- Ce message et toutes les pieces jointes (ci-apres le "message") sont etablis a l'intention exclusive de ses destinataires et sont confidentiels. Si vous recevez ce message par erreur, merci de le detruire et d'en avertir immediatement l'expediteur. Toute utilisation de ce message non conforme a sa destination, toute diffusion ou toute publication, totale ou partielle, est interdite, sauf autorisation expresse. L'internet ne permettant pas d'assurer l'integrite de ce message, BNP PARIBAS (et ses filiales) decline(nt) toute responsabilite au titre de ce message, dans l'hypothese ou il aurait ete modifie. N'imprimez ce message que si necessaire, pensez a l'environnement.

2 1

astype
by Nils Wagner Feb. 14, 2009

Feb. 14, 2009

Hi all, How can I convert an array with string elements to an array with float entries ? >>> coord_info[:,1] array(['0,0', '100,0', '200,0', '300,0', '400,0', '500,0', '600,0', '700,0', '800,0', '0.0', '100.0', '200.0', '300.0', '400.0', '500.0', '600.0', '700.0', '800.0', '0.0', '100.0', '200.0', '300.0', '400.0', '500.0', '600.0', '700.0', '800.0', '0.0', '100.0', '200.0', '300.0', '400.0', '500.0', '600.0', '700.0', '800.0', '0.0', '100.0', '200.0', '300.0', '400.0', '500.0', '600.0', '700.0', '800.0'], dtype='|S50') >>> coord_info[:,1].astype(float) Traceback (most recent call last): File "<stdin>", line 1, in <module> ValueError: invalid literal for float(): 0,0 Nils

2 2

Float and locale formatting fixes merged in trunk
by David Cournapeau Feb. 14, 2009

Feb. 14, 2009

Hi, I've just merged the branch Pauli and me have been working on recently to fix several local and float formatting bugs. All tests pass on Linux, but there may still be rough edges on Windows. Let me know if this causes trouble, cheers, David

5 7

Purpose for bit-wise and'ing the initial mersenne twister key?
by Michael S. Gilbert Feb. 14, 2009

Feb. 14, 2009

In numpy/random/mtrand/randomkit.c on line 159, the initial mersenne twister key (populated from /dev/urandom) gets bit-wise and'ed with 0xffffffff. I'm just curious as why this is done. A bit-wise and with all ones should just give you your original quantity back, right? I don't think there is a problem since the operation doesn't really do anything, and the same thing exists in the mersenne twister reference code, but I am curious as to why it is even there in the first place. Thanks for any thoughts. Regards, Mike

3 19

improvement request to np.dot(a, b) - extended precision summation
by David Henderson Feb. 13, 2009

Feb. 13, 2009

Hello all, I'd like accumulate the summation in extended precision, "double" sum for float inputs, "long double" sum for "double" inputs. A way of doing this is to add an optional third argument to dot - to specify the summation type. I've looked into the code, and the source file multiarraymodule.c has a routine: new_array_for_sum(PyArrayObject *ap1, PyArrayObject *ap2, If the implementation is to pass a third argument to dot that specifies the resultant type, this routine would be modified to add that third argument or type in its priority selection logic. I've not written any code yet, and I'm not fussy as to how it gets done. Is there another way to do this that I'm missing? Thanks in advance, David

2 1

Integer cast problems
by Ralph Kube Feb. 13, 2009

Feb. 13, 2009

Hi there, I have a little problem here with array indexing, hope you see the problem. I use the following loop to calculate some integrals import numpy as N from scipy.integrate import quad T = 1 dt = 0.005 L = 3 n = 2 ints = N.zeros([T/dt]) for t in N.arange(0, T, dt): a = quad(lambda x:-1*(1-4*(t**4))*N.exp(-t**4)*N.exp(-x**2)*N.cos(n*N.pi*(x-L)/(2*L)), -L, L)[0] ints[int(t/dt)] = a print t, N.int32(t/dt), t/dt, a, ints[int(t/dt)] The output from the print statement looks like: 0.14 28 28.0 2.52124867251e-16 2.52124867251e-16 0.145 28 29.0 2.03015199575e-16 2.03015199575e-16 0.15 30 30.0 2.40857836418e-16 2.40857836418e-16 0.155 31 31.0 2.52191011339e-16 2.52191011339e-16 The same happens on the ipython prompt: 0.145 * 0.005 = 28.999999999999996 N.int32(0.145 * 0.005) = 28 Any ideas how to deal with this? Cheers, Ralph

4 4

problem with WRITEABLE flag and array interface
by Nathaniel Smith Feb. 13, 2009

Feb. 13, 2009

I'm using rpy2 to access the R statistical programming runtime from Python, and one of rpy2's nice features is that wrappers for R arrays support the Python array interface (via __array_struct__), so that one can conveniently cast them to numpy arrays and edit them in place. But this seems to expose a bug in numpy's WRITEABLE flag handling. Such arrays start out writeable: >>> a = np.asarray(r) >>> a.flags["WRITEABLE"] True And you can toggle them unwriteable: >>> a.flags["WRITEABLE"] = False But then you are stuck! You cannot make the array writeable again: >>> a.flags["WRITEABLE"] = True ValueError: cannot set WRITEABLE flag to True of this array This may seem rather trivial, but it's completely broken my API design... the problem is that R has somewhat complex rules for when you can write to an array and when you cannot (because of copy-on-write stuff going on behind the scenes), and I wanted to reflect that in my high-level API (for people who don't want to know about R's COW minutiae) by toggling the WRITEABLE flag in a controlled way. But I can't. Help? Self-contained test case: class Fraud(object): pass f = Fraud() a = np.array([1, 2, 3]) f.__array_interface__ = a.__array_interface__ f_asarray = np.asarray(f) assert f_asarray.flags["WRITEABLE"] = True f_asarray.flags["WRITEABLE"] = False f_asarray.flags["WRITEABLE"] = True # Fails -- Nathaniel

1 0

Filling gaps
by A B Feb. 13, 2009

Feb. 13, 2009

Hi, Are there any routines to fill in the gaps in an array. The simplest would be by carrying the last known observation forward. 0,0,10,8,0,0,7,0 0,0,10,8,8,8,7,7 Or by somehow interpolating the missing values based on the previous and next known observations (mean). Thanks.

3 4

umath build failures @ trunk on some platforms?
by Pauli Virtanen Feb. 13, 2009

Feb. 13, 2009

Hi, The Buildbot (up once again) is showing build failures on some platforms: http://buildbot.scipy.org/builders/Windows_XP_x86_64_MSVC/builds/875/steps/… http://buildbot.scipy.org/builders/Linux_SPARC_64_Debian/builds/423/steps/s… Are these a sign of some bug (in the new math config system?), or a glitch in the buildslaves only? -- Pauli Virtanen

3 2