Mailman 3 January 2010 - NumPy-Discussion

is there any alternative to savefig?
by Robert Kiwanuka 28 Jan '10

28 Jan '10

Hi all, I wonder if anyone knows any alternative function in pylab (or otherwise) that could be used to save an image. My problem is as follows: --------------- from pylab import * ... figure(1) fig1 = gca() figure(2) fig2 = gca() figure(3) fig3 = gca() for i,data_file in enumerate(data_file_list): time,x, y,x2, y2 = read_csv_file_4(open (data_file),elements=num_of_ elements) fig1.plot(-x,-y,color=colours[i],label=labellist[i]) fig2.plot(time,-y,color=colours[i],label=labellist[i]) fig3.plot(time,-x,color=colours[i],label=labellist[i]) fig1.legend(loc='best') fig1.set_title("y1 - x1") fig1.set_ylabel("y1") fig1.set_xlabel("x1") #savefig("y1-x1.png") fig2.legend(loc='best') fig2.set_title("y1 - time") fig2.set_ylabel("y1") fig2.set_xlabel("time[s]") #savefig("y1-time.png") fig3.legend(loc='best') fig3.set_title("x1 - time") fig3.set_ylabel("x1") fig3.set_xlabel("time[s]") #savefig("x1-time.png") show() --------------------------- In the above code, I read multiple data files and plot three separate figures. Now I would like to save each of the figures to a file as the commented savefig satements suggest. The trouble is that if I uncomment all those savefig statements, I get three saved images all containing the plot belonging to figure(3), which was the last figure declared. I understand this to be happening because savefig will save the "current" figure, which in this case happens to be the last one declared. If I could do something like fig1.savefig("y1-x1.png") or savefig("y1- x1.png").fig1, this would solve the problem but I'm not aware of any such methods or modules to enable this. This is thus a flaw in the general design/implementation of the savefig function, but is there an alternative function to enable me achieve what I need? Is there perhaps a possible tweak to savefig to make it do the same? Thanks in advance, Robert

3 2

Fixing numpy 1.4.0 ABI breakage, and a plea for self-contained, small commits
by David Cournapeau 28 Jan '10

28 Jan '10

Hi, I have investigated further the ABI issues reported for numpy 1.4.0. I can confirm that we have broken the ABI for 1.4.0 compared to 1.3.0 (besides the "trivial" cython issue). The good new is that I have found the issue, the bad news is that I don't know how to (cleanly) fix it. The problem was caused by the new datetime support, in particular the structure PyArray_ArrFuncs which has been modified in a ABI-incompatible way (the first member cast is bigger because NTYPES is bigger). I don't know how to fix this cleanly - the only solution I can see is to to split cast into two parts, the first part the same size as before and the second part at the end of the structure, with a double-case test everytime the cast member is accessed inside the relevant functions.... As an aside, I would like to reiterate my advice for *small* commits. It took me nearly 2 hours to find this because the offending commit was > 4000 LOC, and it would have been very easy to find this were the code committed as a set of small self-contained commits. thanks, David

5 15

glumpy, fast opengl visualization
by Nicolas Rougier 27 Jan '10

27 Jan '10

Hello, This is an update about glumpy, a fast-OpenGL based numpy visualization. I modified the code such that the only dependencies are PyOpenGL and IPython (for interactive sessions). You will also need matplotlib and scipy for some demos. Sources: hg clone http://glumpy.googlecode.com/hg/ glumpy No installation required, you can run all demos inplace. Homepage: http://code.google.com/p/glumpy/ Nicolas

4 3

np.ma.apply_along_axis mask propagation
by Angus McMorland 27 Jan '10

27 Jan '10

Hi all, I'm having an issue with np.ma.apply_along_axis not propagating a mask correctly. This code-snippet should hopefully show what's happening: -------------------- import numpy as np xy = np.random.random(size=(5,2)) mask = np.tile(np.array([True] * 3 + [False] * 2)[:,None], (1,2)) xyma = np.ma.array(xy, mask=mask) def myfunc(vec): x,y = vec return x,y xyma2 = np.ma.apply_along_axis(myfunc, 1, xyma) tst = "np.all(asarray(myfunc(xyma[1])).mask == xyma2[1].mask)" print tst, ":", eval(tst) -> np.all(asarray(myfunc(xyma[1])).mask == xyma2[1].mask) : False -------------------- The point here is not that xyma.mask != xyma2.mask, but that xyma2.mask != the output of myfunc run on the individual rows, which it seems like it should. The following simple change seems to fix this. --- /usr/lib/python2.6/dist-packages/numpy/ma/extras.py 2009-04-05 04:09:20.000000000 -0400 +++ tmp/extras.py 2010-01-27 10:45:10.000000000 -0500 @@ -322,7 +322,7 @@ n -= 1 i.put(indlist, ind) j.put(indlist, ind) - res = func1d(arr[tuple(i.tolist())], *args, **kwargs) + res = asarray(func1d(arr[tuple(i.tolist())], *args, **kwargs)) outarr[tuple(flatten_inplace(j.tolist()))] = res dtypes.append(asarray(res).dtype) k += 1 Does this seem like an improvement? I haven't explored performance issues yet, pending sanity check. Thanks, Angus. -- AJC McMorland Post-doctoral research fellow Neurobiology, University of Pittsburgh

1 0

numpy.i
by Ross Harder 26 Jan '10

26 Jan '10

I'm struggling with using some of the macros in numpy.i for my own typemap. The problem is that the arrayobject.h include does not end up in the c wrapper code after swig runs. numpy.i has at the beginning: %{ #ifndef SWIG_FILE_WITH_INIT # define NO_IMPORT_ARRAY #endif #include "stdio.h" #include <numpy/arrayobject.h> %} I tried explicitly putting %{#include "arrayobject.h"%} into the interface file. It ends up after the fragment code in the wrapper so the compiler is complaining. thanks, ross

2 1

is shuffle needlessly slow?
by Alan G Isaac 26 Jan '10

26 Jan '10

Is this a fair test? I expected shuffle to be much faster (no array creation). Alan Isaac >>> import timeit >>> >>> setup = """ ... import numpy as np ... prng = np.random.RandomState() ... N = 10**5 ... indexes = np.arange(N) ... """ >>> >>> print timeit.timeit('prng.shuffle(indexes)',setup, number=100) 5.69172311006 >>> print timeit.timeit('indexes = prng.random_sample(N).argsort()',setup, number=100) 1.54648202495

2 2

Re: [Numpy-discussion] indexing, searchsorting, ...
by Jan Strube 26 Jan '10

26 Jan '10

Dear Josef and Keith, thank you both for your suggestions. I think intersect would be what I want for it makes clean code. I have, however, spotted the problem: I was mistakenly under the assumption that random_integers returns unique entries, which is of course not guaranteed, so that the random sample contained duplicate entries. That's why the numpy methods returned results inconsistent with python 'in'. I'll have to be a bit smarter in the generation of the random sample. Good thing I try to do things in two different ways. (Sometimes it is, anyway...) Thanks again for your quick help. Cheers, Jan

2 2

should numpy.equal raise an exception instead of returning NotImplemented?
by Ernest Adrogué 26 Jan '10

26 Jan '10

Hi, Do you think it is sensible for np.equal to return a NotImplemented object when is given an array of variable length dtype? Consider this code: x = np.array(['xyz','zyx']) np.where(np.equal(x, 'zyx'), [0,0], [1,1]) the last line returns array([0, 0]) which is wrong. Compare with np.where(x == 'zyx', [0,0], [1,1]) I think in this case raising an exception would be better, because of the way np.equal is often used as an argument to another function. It's not easy to see where the bug is or even that there is a bug in your code. Cheers. Ernest

2 2

pgcc-Error-Unknown switch
by A.MARSAN 26 Jan '10

26 Jan '10

Dear All, I'm trying to use f2py in order to convert a fortan-file fonctions_f90.f90 I apply the following command line f2py -m fonctions_f90 --fcompiler=pg -c fonctions_f90.f90 Everything seems to be well, until this error appears : Pgcc: /tmp/tmp7RvKeA/src.linux-x86_64-2.5/fortranobject.c Pgcc-Error-Unknown switch: -fno-strict-aliasing Pgcc-Error-Unkown switch: -fwrapv Pgcc-Error-Unkown switch: -Wall Pgcc-Error-Unkown switch: -Wstrict-prototypes Does anyone could explain me what append and how to fix this problem ? Thanks for help. -- Aurelien

1 0

indexing, searchsorting, ...
by Jan Strube 25 Jan '10

25 Jan '10

Dear List, I'm trying to speed up a piece of code that selects a subsample based on some criteria: Setup: I have two samples, raw and cut. Cut is a pure subset of raw, all elements in cut are also in raw, and cut is derived from raw by applying some cuts. Now I would like to select a random subsample of raw and find out how many are also in cut. In other words, some of those random events pass the cuts, others don't. So in principle I have randomSample = np.random.random_integers(0, len(raw)-1, size=sampleSize) random_that_pass1 = [r for r in raw[randomSample] if r in cut] This is fine (I hope), but slow. I have seen searchsorted mentioned as a possible way to speed this up. Now it gets complicated. I'm creating a boolean array that contains True, wherever a raw event is in cut. raw_sorted = np.sort(raw) cut_sorted = np.sort(cut) passed = np.searchsorted(raw_sorted, cut_sorted) raw_bool = np.zeros(len(raw), dtype='bool') raw_bool[passed] = True Now I create a second boolean array that is set to True at the random values. The events I care about are the ones that pass the cuts and are selected by the random selection: sample_bool = np.zeros(len(raw), dtype='bool') sample_bool[randomSample] = True random_that_pass2 = raw[np.logical_and(raw_bool, sample_bool)] The problem comes in now: random_that_pass2 and random_that_pass1 have different lengths!!! Sometimes one is longer, sometimes the other. I am completely at a loss to explain this. I tend to believe the slow selection leading to random_that_pass1, because it's only two lines, but I don't understand where the other selection could fail. Unfortunately, the samples that give me trouble are 2.2 MB, so maybe a bit large to mail around, but I can put it somewhere if needed. Thank you for your help, Cheers, Jan

3 2