Mailman 3 January 2008 - NumPy-Discussion

numscons, available as python egg
by David Cournapeau 25 Jan '08

25 Jan '08

Hi, Sorry for the flooding, but I finally managed to build an egg and put it on the web, so numscons is available as an egg, now. You should be able to install it using easy_install, e.g easy_install numscons should work. cheers, David

1 0

Re: [Numpy-discussion] How to build on Solaris 10 (x86) using sunperf?
by Peter Ward 25 Jan '08

25 Jan '08

David Cournapeau wrote: > The current config.h works fine for solaris with Sun compilers, in my > experience, so the problem must be somewhere else. > > Peter, could you post the errors you got ? As an alternative, I am > working on an alternative build system for numpy: it should work on > solaris (tested on Indiana, and other people reported success on > solaris > 9 and 10). Unfortunately, I have been redesigning the internals > quite a > lot lately, and I have not tested the changes on solaris yet. If > you are > willing to test, it should be easy for me to make it work again on > solaris in a few minutes, though. > The problems I was having were due to a bad site.cfg initially and then a problem with the python pkg from sunfreeware (ctypes version mismatch). Numpy is now happily installed. If you need someone to test anything new, let me know. Thanks! Pete

2 1

Numpy Example List with Doc updates
by Filip Wasilewski 25 Jan '08

25 Jan '08

Hi all, Just a small note. I've updated the `Numpy Example List With Doc`[1] that is linked from the main doc index. Because I'm not always in the loop and the page likes to get out of sync with the base `Numpy Example List`[2] and NumPy, I've posted some instructions[3] that you might find useful when trying to bring it up-to-date :) [1] http://scipy.org/Numpy_Example_List_With_Doc [2] http://scipy.org/Numpy_Example_List [3] http://scipy.org/Numpy_Example_List_With_Doc/script cheers, f -- http://filipwasilewski.pl

2 1

Docstring standard: how to specify variable types
by Stefan van der Walt 25 Jan '08

25 Jan '08

Hi all, The numpy documentation standard example shows: Parameters ---------- var1 : array_like Array_like means all those objects -- lists, nested lists, etc. -- that can be converted to an array. var2 : integer Write out the full type long_variable_name : {'hi', 'ho'}, optional Choices in brackets, default first when optional. I'd like to know: 1. "array_like" describes objects that can be forced to quack like ndarrays. Are there any other such "special" descriptions? 2. How do we specify default values? 3. Why do we need the "optional" keyword (the function signature already indicates that the parameter is optional). 4. Do we really need the "Other Parameters" list? It would make more sense to split positional and keyword arguments, but I'm not even sure that is necessary, since that information is already specified in the function signature. 5. Is the {'hi', 'ho'} syntax used when a parameter can only assume a limited number of values? In Python {} is a dictionary, so why not use ('hi','ho') instead? Thanks for your feedback! Regards Stéfan

4 7

numpy.ma.compress
by Eric Firing 24 Jan '08

24 Jan '08

Pierre, numpy.compress exists, but numpy.ma.compress does not; is this intentional? Eric

6 13

numpy.concatenate doesn't check axis for 1D case. Bug or feature?
by Stuart Brorson 24 Jan '08

24 Jan '08

As the subject says, numpy.concatenate doesn't seem to obey -- or even check -- the axis flag when concatenating 1D arrays: ----------------------- <session log> ----------------- In [30]: A = numpy.array([1, 2, 3, 4]) In [31]: D = numpy.array([6, 7, 8, 9]) In [32]: numpy.concatenate((A, D)) Out[32]: array([1, 2, 3, 4, 6, 7, 8, 9]) In [33]: numpy.concatenate((A, D), axis=0) Out[33]: array([1, 2, 3, 4, 6, 7, 8, 9]) In [34]: numpy.concatenate((A, D), axis=1) Out[34]: array([1, 2, 3, 4, 6, 7, 8, 9]) In [35]: numpy.concatenate((A, D), axis=2) Out[35]: array([1, 2, 3, 4, 6, 7, 8, 9]) ----------------------- </session log> ----------------- However, if you create the same arrays as 2D (1xn) arrays, then numpy checks, and does the right thing: ---------------------- <session log> ------------------- In [36]: A = numpy.array([[1, 2, 3, 4]]) In [37]: D = numpy.array([[6, 7, 8, 9]]) In [38]: A.shape Out[38]: (1, 4) In [39]: numpy.concatenate((A, D)) Out[39]: array([[1, 2, 3, 4], [6, 7, 8, 9]]) In [40]: numpy.concatenate((A, D), axis=0) Out[40]: array([[1, 2, 3, 4], [6, 7, 8, 9]]) In [41]: numpy.concatenate((A, D), axis=1) Out[41]: array([[1, 2, 3, 4, 6, 7, 8, 9]]) In [42]: numpy.concatenate((A, D), axis=2) --------------------------------------------------------------------------- <type 'exceptions.ValueError'> Traceback (most recent call last) /fs/home/sdb/<ipython console> in <module>() <type 'exceptions.ValueError'>: bad axis1 argument to swapaxes ----------------------- </session log> ----------------- Question: Is is a bug or a feature? I'd at least think that numpy would check the axis arg in the 1D case, and issue an error if the user tried to do something impossible (e.g. axis=2). Whether numpy would promote a 1D to a 2D array is a different question, and I am agnostic about that one. Cheers, Stuart Brorson Interactive Supercomputing, inc. 135 Beaver Street | Waltham | MA | 02452 | USA http://www.interactivesupercomputing.com/

1 0

Re: [Numpy-discussion] numpy.ma.compress
by Eric Firing 23 Jan '08

23 Jan '08

Pierre GM wrote: > On Wednesday 23 January 2008 16:17:51 you wrote: >> Pierre, >> >> numpy.compress exists, but numpy.ma.compress does not; is this intentional? > > Probably not. I usually don't use this function, preferring to use indexing > instead. If you have a need for it, I can probably come up with something > relatively soon: the basis would be to apply compress first on the ._data > part, return a view with the same class as the original object, and update > the mask with compress as needed. The only reason to add it would be backwards compatibility. Evidently it exists in original numpy.ma. Mike D used it in mpl's path.py until today when someone pointed out that it did not work with the maskedarray branch. (I think I tripped over the same thing a couple days ago, but worked around it at a higher level.) I agree that it is better *not* to use it, and we can easily strip it out of mpl if it occurs anywhere else; but there may be other user code that will trip over its absence when 1.05 comes out, so it might be easier to put it in, preserving similarity to numpy as well as old numpy.ma, than to leave it out and have to field future questions about it. Eric

1 0

Sage/Scipy Days 8 reminder: Feb 29-March 4.
by Fernando Perez 23 Jan '08

23 Jan '08

Hi all, Just a quick reminder for all about the upcoming Sage/Scipy Days 8 at Enthought collaborative meeting: http://wiki.sagemath.org/days8 Email me directly (Fernando.Perez(a)Colorado.edu) if you plan on coming, so we can have a proper count and plan accordingly. Cheers, f

2 2

__array_interface__ / __array_struct__
by Thomas Heller 23 Jan '08

23 Jan '08

I am experimenting with implementing __array_interface__ and/or __array_struct__ properties for ctypes instances, and have problems to create numpy arrays from them that share the memory. Probably I'm doing something wrong; what is the correct function in numpy to create these shared objects? I am using numpy.core.multiarray.array(ctypes-object), is that correct? Thanks, Thomas

2 4

changed behavior of numpy.histogram
by Mark.Miller 23 Jan '08

23 Jan '08

Greetings: I just noticed a changed behavior of numpy.histogram. I think that a recent 'fix' to the code has changed my ability to use that function (albeit in an unconventional manner). I previously used the histogram function to obtain counts of each unique string within a string array. Again, I recognize that it is not a typical use of the histogram function, but it did work very nicely for me. Here's an example: ###numpy 1.0.3 --works just fine >>> import numpy >>> numpy.__version__ '1.0.3' >>> a=numpy.array(('atcg', 'atcg', 'aaaa', 'aaaa')) >>> a array(['atcg', 'atcg', 'aaaa', 'aaaa'], dtype='|S4') >>> b=numpy.unique(a) >>> numpy.histogram(a,b) (array([2, 2]), array(['aaaa', 'atcg'], dtype='|S4')) >>> ###numpy 1.0.4 --no longer functions >>> import numpy >>> numpy.__version__ '1.0.4' >>> a=numpy.array(('atcg', 'atcg', 'aaaa', 'aaaa')) >>> a array(['atcg', 'atcg', 'aaaa', 'aaaa'], dtype='|S4') >>> b=numpy.unique(a) >>> numpy.histogram(a,b) Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/opt/libraries/python/python-2.5.1/numpy-1.0.4-gnu/lib/python2.5/site-packages/numpy/lib/function_base.py", line 154, in histogram if(any(bins[1:]-bins[:-1] < 0)): TypeError: unsupported operand type(s) for -: 'numpy.ndarray' and 'numpy.ndarray' >>> Is this something that can possibly be fixed (should I submit a ticket)? Or should I revert to some other approaches for implementing the same idea? It really was a nice convenience. Or, alternately, would some sort of new function along the lines of a numpy.countunique() ultimately be useful? Thanks, -Mark

3 4