Mailman 3 July 2013 - SciPy-Dev

subclassing scipy.stats distributions - request for feedback
by Ralf Gommers July 10, 2013

July 10, 2013

Hi all, We're very close to merging https://github.com/scipy/scipy/pull/2588, which is a nice enhancement by Evgeni Burovski that allows specifying shape parameters as keyword arguments for all distribution methods. With this kind of change it's hard to verify that we're not breaking any non-standard subclasses that people have written. Therefore I have the following requests: 1. please test your code with that PR 2. please provide code samples or descriptions of the way you subclass distributions (if different from what's done in scipy itself), to be used to augment the test suite and documentation. Thanks, Ralf

1 0

`a` and `b` attributes of the stats distributions
by Warren Weckesser July 9, 2013

July 9, 2013

I've been working on a pull request for one of the distributions in the stats module, and as often happens, this has lead to the urge to do some more clean up. The distributions defined in stats/distributions.py have attributes `a` and `b` that give the lower and upper ends of the support of the distribution (i.e. outside of [a, b], the PDF is 0). A problem with this API is that for many distributions, the support depends on the parameters. Currently, this is handled in many of the distributions by modifying self.a and self.b in the _argcheck() method (I guess under the assumption that any method that needs `a` and `b` will have called `_argcheck()` at some point). This leads to stateful behavior such as: In [1]: from scipy.stats import genpareto In [2]: genpareto.b # Initially b is inf. Out[2]: inf In [3]: genpareto.cdf(0, -2) Out[3]: 0.0 In [4]: genpareto.b # Now b is 0.5 (as a scalar array) Out[4]: array(0.5) In [5]: genpareto.cdf(0, 1) Out[5]: 0.0 In [6]: genpareto.b Out[6]: array(inf) # b is back to inf, but as a scalar array. This API is ugly, and I'd like to fix it. To start, I think there should be a method `_support`: def _support(self, *args): """Returns a tuple (a,b) that gives the support of the distribution.""" ... All the distributions would be modified to call `self._support(...)` to get `a` and `b` instead of using the attributes `self.a` and `self.b`. (The method could be public--I just defaulted to private while working out the appropriate API.) It would be nice to get rid of `a` and `b` entirely, but they are currently public attributes, so I don't think we can simply remove them. Comments? Suggestions? Warren

3 4

AttributeError when editing source
by Anubhab Baksi July 7, 2013

July 7, 2013

Hi, I was trying to write an additional function in misc/common. But, when I tried to run that, I got an AttributeError, stating module object has no such attribute. Can anybody, please, tell me why this is happening and how can I fix it? Thank you all.

2 1

Compiling scipy on AIX 6.1
by Bruno Luciano Amadio Caires July 5, 2013

July 5, 2013

Hi, I am trying install scipy on AIX 6.1 power7, but I'm getting errors. I am using XL compiler, and scipy-0.12.0/scipy-0.8.0 I have numpy-1.7.0 working that was compiling here with XL. Has someone a HOWTO? C compiler: xlc_r -ma -I/opt/freeware/include -DAIX_GENUINE_CPLUSCPLUS -Wl,-brtl -qmaxmem=16384 -DSYSV -D_AIX -D_AIX32 -D_AIX41 -D_AIX43 -D_AIX51 -D_ALL_SOURCE -DFUNCPROTO=15 -O -I/opt/freeware/include -DNDEBUG -O compile options: '-DNO_ATLAS_INFO=1 -I/opt/freeware/lib/python2.6/site-packages/numpy/core/include -I/opt/freeware/include/python2.6 -c' xlc_r: scipy/integrate/_odepackmodule.c /opt/freeware/lib/python2.6/config/ld_so_aix /usr/bin/xlf95 -bI:/opt/freeware/lib/python2.6/config/python.exp -bshared -F/tmp/tmpcqh5Ki/dcxixc_xlf.cfg build/temp.aix-6.1-2.6/scipy/integrate/_odepackmodule.o -L/home/bruno/local/lib -Lbuild/temp.aix-6.1-2.6 -lodepack -llinpack_lite -lmach -lblas -o build/lib.aix-6.1-2.6/scipy/integrate/_odepack.so ld: 0711-224 WARNING: Duplicate symbol: guesses ld: 0711-345 Use the -bloadmap or -bnoquiet option to obtain more information. ld: 0711-317 ERROR: Undefined symbol: .Py_InitModule4_64 ld: 0711-317 ERROR: Undefined symbol: .daxpy_ ld: 0711-317 ERROR: Undefined symbol: .ddot_ ld: 0711-317 ERROR: Undefined symbol: .idamax_ ld: 0711-317 ERROR: Undefined symbol: .dscal_ ld: 0711-224 WARNING: Duplicate symbol: guesses ld: 0711-345 Use the -bloadmap or -bnoquiet option to obtain more information. ld: 0711-317 ERROR: Undefined symbol: .Py_InitModule4_64 ld: 0711-317 ERROR: Undefined symbol: .daxpy_ ld: 0711-317 ERROR: Undefined symbol: .ddot_ ld: 0711-317 ERROR: Undefined symbol: .idamax_ ld: 0711-317 ERROR: Undefined symbol: .dscal_ error: Command "/opt/freeware/lib/python2.6/config/ld_so_aix /usr/bin/xlf95 -bI:/opt/freeware/lib/python2.6/config/python.exp -bshared -F/tmp/tmpcqh5Ki/dcxixc_xlf.cfg build/temp.aix-6.1-2.6/scipy/integrate/_odepackmodule.o -L/home/bruno/local/lib -Lbuild/temp.aix-6.1-2.6 -lodepack -llinpack_lite -lmach -lblas -o build/lib.aix-6.1-2.6/scipy/integrate/_odepack.so" failed with exit status 8 att -- Bruno L. Amadio Caires Informática - FT Universidade Estadual de Campinas Fone:(19)2113-3470

2 3

Missing statistical Tests for scipy.stats
by Haslwanter Thomas July 5, 2013

July 5, 2013

Hi, in preparing a statistics lecture using Python (http://work.thaslwanter.at/Stats/html/ ), I realized that two important statistical tests are currently not available in scipy: · McNemar Test (http://en.wikipedia.org/wiki/McNemar%27s_test ) · Cochran's Q-Test (http://en.wikipedia.org/wiki/Cochran%27s_Q_test) I have found the equations for the McNemar test, and implemented a version in Python. Would there be an interest in incorporating this module in scipy.stats? If I'd get the formulas for Cochran's Q-test, I would also be keen on implementing that one. What do you think? thomas --- Prof. (FH) PD Dr. Thomas Haslwanter School of Applied Health and Social Sciences University of Applied Sciences Upper Austria FH OÖ Studienbetriebs GmbH Garnisonstraße 21 4020 Linz/Austria Tel.: +43 (0)5 0804 -52170 Fax: +43 (0)5 0804 -52171 E-Mail: Thomas.Haslwanter(a)fh-linz.at<mailto:Thomas.Haslwanter@fh-linz.at> Web: me-research.fh-linz.at<http://work.thaslwanter.at> or work.thaslwanter.at

3 3

Migrating to Markdown Syntax from reStructured Text in SPC
by Surya Kasturi July 5, 2013

July 5, 2013

Hi all, its just an idea and appreciate your comments. I am thinking about migrating text markup that spc currently supports from reStructured Text to Markdown.. Reasons: 1. reST has some security issues - Someone can put some raw HTML in it and thus creating vulnerabilities.. Better explained at https://docs.djangoproject.com/en/1.4/ref/contrib/markup/#restructured-text 2. currently reST to HTML conversion is done on the server side! It could be done on client side enabling better text editing and previewing.. Markdown seems to be a decent option since there are some opensource tools (by Stackoverflow) and we can start integrating them with our site http://code.google.com/p/pagedown/wiki/PageDown (link previously referred by Sebastian during discussion - thanks!) Eventhough we might not move to Markdown, there seems to be some tools django offers for handling text markup https://docs.djangoproject.com/en/1.4/ref/contrib/markup/ And we dont have to use convert_rest_to_html() at all!! Probably, we might need to convert everytime the user loads the page (have to look into the exact docs.. don't know) but it offers us another feature Rightnow, the image uploads embedded in description as hardcoded during conversion i.e., the URL looks something like " scipy-central.org/media/blah-blah/122.jpg" If for some reasons we change the domain name or want to work on localhost, these images don't work! https://github.com/scipy/SciPyCentral/issues/123 It would be nice to have your views.. its not part of GSoC but probably, we can put it in..or if I got some extra time, I can do without mentioning in GSoC proposal. Thanks Surya

4 6

GSoC -- numpy interactions with sparse matrices
by Blake Griffith July 2, 2013

July 2, 2013

Hello scipy, Today marks the first day of the next phase of my GSoC proposal, numpy interactions and overriding ufuncs for sparse matrices. I've been thinking about how I will handle the ufunc interactions. I'm still considering how to implement a ufunc override mechanism. However I think the existing functionality in numpy, (__array_prepare__, __array_finalize__, etc) would be enough for a few cases: * unary element-wise ufuncs. Because these could be applied to the .data attribute using array prepare. * binary element-wise ufuncs that correspond to python operators (__add__, __gt__, __mul__, etc.) because there is already functionality in place to override these. This of course leave out: logaddexp logaddexp2 ones_like arctan2 hypot maximum minimum copysign nextafter ldexp (However I think it would be possible to override these using the existing functionality for overriding python overrides but this would be a bit hackish, and numpy might not like the changes.) None of these left over ufuncs seem to be the most widely used (I could be wrong). So going about this without adding an ufunc override attribute would still add a lot of functionality. So I think the best course would be to explore adding a override attribute but if it is extremely difficult I can resort to this. comment, suggestions, criticisms?

3 2

Quadrature of a vector-valued function, implementation changing the quadpack package
by Saullo Castro July 2, 2013

July 2, 2013

Dear all, I am implementing a vectorized version of: `dagse.f` --> `dagsev.f` and `dqk21.f` --> `dqk21v.f` Which handles the vectorized integration in "low level" inside the Fortran routines. I would like to discuss with the SciPy dev group if there is any interest to make this feature part of SciPy. Also, I need some advice about how to wrap this with quadpack. Currently I am changing: `_quadpackmodule.c` `_quadpack.h` to make `dagsev.f` work similarly to `dagse.f`. Thank you! Saullo

2 1