Mailman 3 August 2009 - NumPy-Discussion

future directions
by Fons Adriaensen 28 Aug '09

28 Aug '09

Some weeks ago there was a post on this list requesting feedback on possible future directions for numpy. As I was quite busy at that time I'll reply to it now. My POV is that of a novice user, who at the same time wants quite badly to use the numpy framework for his numerical work which in this case is related to (some rather advanced) multichannell audio processing. >From that POV, I'd suggest the following: 1. Adopt an object based on Python-3's buffer protocol as the basic array type. It's immensely more powerful than ndarray, while at the same time it's close enough to ndarray to allow a gradual adoption. 2. Adopting that format will make it even more important to clearly define in which cases data gets copied and when not. This should be based on some simple rules that can be evaluated by a code author without requiring a lookup in the reference docs each time. 3. Finally remove all the redundancy and legacy stuff from the world of numerical Python. It is *very* confusing to a new user. 4. Ensure that each package deals with one problem area only. For example a package that (by its name) suggests it provides plotting facilities should provide only plotting facilities, and not spectra, averages of all sorts, etc. 5. Ensure some consistency in style. Some numerical Python packages use two-character function names, some a have veryLongCamelCased names. Just my two Eurocents of course. Ciao, -- FA Io lo dico sempre: l'Italia è troppo stretta e lunga.

6 6

Iterate over an array
by davide lasagna 28 Aug '09

28 Aug '09

Hi all, I ve got a 2d array and i want to iterate over its columns in a "pythonic way". This is what i have in mind: please consider this snippet: ################################################# import numpy as np array = np.random.standard_normal( (10,10) ) for column in array.some_column_method(): column = do_something() ################################################# The trivial way do the for loop is: ################################################# for i in range( array.shape[1] ): array[:, i] = do_something() ################################################# Is there any way to do what i think?? Can i obtain "pythonically" a list of column arrays?? Any help is appreciated. Cheers.. Davide Lasagna Dip. Ingegneria Aerospaziale Politecnico di Torino Italia

2 1

Re: [Numpy-discussion] future directions
by Joe Harrington 28 Aug '09

28 Aug '09

Christopher Barker wrote: >> Following the full >> PEP procedure >or a parallel NPEP system. Actually, I originally intended just to mean "follow the procedure" not "do it in their system". But, in thinking about it, if it's compatible with their system to develop a whole subpackage in their procedure space, we should. Ultimately the decision to include something in Python is one they will make, and such decisions are largely social ones. The more they have seen, been included in, and had chance to comment on our stuff, the more bought in they are and the less chance there is of objection to it. As for rapid pace of change, it would be feasible to agree on a core set of functionality that should go in the main language, to look at that with our decade of experience, and decide on a design and implementation that would be stable relative to what we have now. Mostly the code would just transfer; what people complain about is fluff, underlying package structure, API inconsistency, and turds. A few things might have API changes (like np.median() not long ago). There would be some significant issues to decide to do or not to do (like generalized ufuncs not long ago), but not many. I am thinking small here, though others will differ. No financials, nothing now deprecated, no fromnumeric namespace, nothing that is likely to evolve fast. Simple, clean, organized. The rest goes into scipy. --jh--

1 0

What type should / return in python 3k when applied to two integer types?
by Charles R Harris 28 Aug '09

28 Aug '09

I'm thinking double. There is a potential loss of precision for 64 bit ints but nothing else seems reasonable for a default. Thoughts? Chuck

11 35

a[j, k], clipping j k to the edge if they're 1 off ?
by denis bzowy 28 Aug '09

28 Aug '09

Folks, I want to index a[j,k], clipping j or k to the edge if they're 1 off -- def aget( a, j, k ): """ -> a[j,k] or a[edge] """ # try: # return a[j,k] -- nope, -1 # except IndexError: m,n = a.shape return a[ min(max(j, 0), m-1), min(max(k, 0), n-1)] This works but is both ugly and 5* slower than plain a[j][k]. Is there a better way ? (Sorry if this is a duplicate, must come up often.) cheers -- denis

2 1

c++ comments in parse_datetime.c
by Charles R Harris 28 Aug '09

28 Aug '09

Hi Travis, The new parse_datetime.c file contains a lot of c++ style comments that should be fixed. Also, the new test for mirr is failing on all the buildbots. Chuck

4 7

[ANN] PyOpenCL 0.90 - a Python interface for OpenCL
by Andreas Klöckner 28 Aug '09

28 Aug '09

What is it? ----------- PyOpenCL makes the industry-standard OpenCL compute abstraction available from Python. PyOpenCL has been tested to work with AMD's and Nvidia's OpenCL implementations and allows complete access to all features of the standard, from a nice, Pythonic interface. Where can I get it? ------------------- Homepage: http://mathema.tician.de/software/pyopencl Download: http://pypi.python.org/pypi/pyopencl Documentation: http://documen.tician.de/pyopencl Wiki: http://wiki.tiker.net/PyOpenCL Main Features ------------- * Object cleanup tied to lifetime of objects. This idiom, often called RAII in C++, makes it much easier to write correct, leak- and crash-free code. * Completeness. PyOpenCL puts the full power of OpenCL’s API at your disposal, if you wish. Every obscure get_info() query and all CL calls are accessible. * Automatic Error Checking. All errors are automatically translated into Python exceptions. * Speed. PyOpenCL’s base layer is written in C++, so all the niceties above are virtually free. * Helpful, complete Documentation If that sounds similar to PyOpenCL's sister project PyCUDA [1], that is not entirely a coincidence. :) License ------- PyOpenCL is open-source under the MIT/X11 license and free for commercial, academic, and private use. Andreas [1] http://mathema.tician.de/software/pycuda

1 0

Fortran reader for npy files
by David Huard 28 Aug '09

28 Aug '09

Hi, Has someone written a fortran reader for the "npy" binary files numpy.save creates ? Thanks, David

1 0

Performance testing in unit tests
by Gael Varoquaux 28 Aug '09

28 Aug '09

Hi list, This is slightly off topic, so please pardon me. I want to do performance testing. To be precise, I have a simple case: I want to check that 2 operations perform with a similar speed (so I am abstracted from the machines performance). What would be the recommended way of timing the operation in a unit test? I use nose, if this is of any use. I am more than happy to be pointed to an example. Cheers, Ga�l

2 3

histogram: sum up values in each bin
by Tim Michelsen 27 Aug '09

27 Aug '09

Hello, I need some advice on histograms. If I interpret the documentation [1, 2] for numpy.histogram correctly, the result of the function is a count of the occurences sorted into each bin. (n, bins) = numpy.histogram(v, bins=50, normed=1) But how can I apply another function on these values stacked in each bin? Like summing them up or building averages? Thanks, Timmie [1] http://docs.scipy.org/doc/numpy/reference/generated/numpy.histogram.html [2] http://www.scipy.org/Tentative_NumPy_Tutorial#head-aa75ec76530ff51a2e98071a…

4 7