Mailman 3 February 2009 - NumPy-Discussion

Tickets for closing
by Pauli Virtanen Feb. 28, 2009

Feb. 28, 2009

Hi, Could someone with the proper Trac permissions close the following tickets: http://scipy.org/scipy/numpy/ticket/884 http://scipy.org/scipy/numpy/ticket/902 Fixed in trunk, when fix_float_format was merged back. http://scipy.org/scipy/numpy/ticket/955 Fixed in trunk + tests. http://scipy.org/scipy/numpy/changeset/6283 Addressed in trunk? Also, the following could be checked/reviewed: http://scipy.org/scipy/numpy/ticket/975 Possibly also fixed? http://scipy.org/scipy/numpy/ticket/629 Does the fix_float_format branch address printing longdouble nans (which I guess was the last issue)? Could the currently commented-out test be uncommented, cf. at the end of numpy/core/tests/test_scalarmath.py? Is float32 still a problem? (The test passes for me.) -- Pauli Virtanen

2 2

Re: [Numpy-discussion] Speedup creation of a 3-color array from a 2-d color-index array a color lut
by Delbert Franz Feb. 27, 2009

Feb. 27, 2009

> Message: 2 > Date: Thu, 26 Feb 2009 23:32:35 -0600 > From: Robert Kern <robert.kern(a)gmail.com> > Subject: Re: [Numpy-discussion] Speedup creation of a 3-color array > from a 2-d color-index array a color lut > To: Discussion of Numerical Python <numpy-discussion(a)scipy.org> > Message-ID: > <3d375d730902262132k5a011d14t84317458f9b6cf87(a)mail.gmail.com> > Content-Type: text/plain; charset=UTF-8 > > On Thu, Feb 26, 2009 at 23:27, Delbert Franz <ddf(a)iqdotdt.com> wrote: > > > > I have ?geotiff files of scanned paper maps that use an indexed color scheme > > with a 256-element > > color lookup table (color lut) and a 9252 by 7420 array ?of uint8 elements. > > The color is given by > > three values. ?I want to create an array with shape: (9252, 7420, 3) so that > > I can display the > > image without creating internal array working space in Matplotlib that > > exeeds 2^31 bytes. > > > > The following three approaches work in that the correct image is displayed, > > but all of them > > are waaaaay too slow:) > > > > ?Let > > ? doq have shape (9252, 7420) and have uint8 elements > > ? ctab have shape (256, 3) and have uint8 elements. > > ? doqq have shape (9252, 7420, 3) and have unit8 elements > > doqq = ctab[doq] > Thanks, Robert! It works just fine but it "blows my mind":) Never would of thought of trying that. I'll have to work hard at getting my "mind around" how Numpy "views the universe". I'm just starting to come to terms with what "object oriented" means in Python. In 46 years of developing software, one is always on a learning curve, sometimes relatively flat, but sometimes nearly vertical--this was a moment on the vertical:) Delbert

2 1

ValueError: invalid literal for float()
by Nils Wagner Feb. 27, 2009

Feb. 27, 2009

Hi all, Is it possible to modify the behaviour of float wrt the following situation >>> permas_M[0,2] '1.5699999809265137D+01' >>> float(permas_M[0,2]) Traceback (most recent call last): File "<stdin>", line 1, in <module> ValueError: invalid literal for float(): 1.5699999809265137D+01 The following works. >>> permas_M[0,2] = '1.5699999809265137E+01' >>> permas_M[0,2] '1.5699999809265137E+01' >>> float(permas_M[0,2]) 15.699999809265137 Nils

2 1

Speedup creation of a 3-color array from a 2-d color-index array a color lut
by Delbert Franz Feb. 27, 2009

Feb. 27, 2009

I have geotiff files of scanned paper maps that use an indexed color scheme with a 256-element color lookup table (color lut) and a 9252 by 7420 array of uint8 elements. The color is given by three values. I want to create an array with shape: (9252, 7420, 3) so that I can display the image without creating internal array working space in Matplotlib that exeeds 2^31 bytes. The following three approaches work in that the correct image is displayed, but all of them are waaaaay too slow:) Let doq have shape (9252, 7420) and have uint8 elements ctab have shape (256, 3) and have uint8 elements. doqq have shape (9252, 7420, 3) and have unit8 elements #1 The way it would be done in a compiled language, like Fortran where it would run in a small fraction of a second instead of several minutes:) But what I want to ultimately accomplish is hard to do in Fortran! for i in range(9252): for j in range(7420): for k in range(3): doqq[i,j,k] = ctab[doq[i,j],k] #2 Try to use some special numpy features: for i in doq.flat: doqq.flat = ctab[i,0:3] #3 Variation of #1 for i in range(9252): for j in range(7420): doqq[i,j,] = ctab[doq[i,j],] All of these use too many element accesses, which are slow. My searching and reading the documentation has not given me an approach that avoids these low-level and slow (in python/numpy) accesses. I suspect there is a clever way to do this but as a newbie I'm having some rough going on getting this process to be much faster. Thanks, Delbert Franz -- View this message in context: http://www.nabble.com/Speedup-creation-of-a-3-color-array-from-a-2-d-color-… Sent from the Numpy-discussion mailing list archive at Nabble.com.

2 1

casting integers to reals
by Gideon Simpson Feb. 26, 2009

Feb. 26, 2009

I want to do: numpy.float(numpy.arange(0, 10)) but get the error: Traceback (most recent call last): File "<stdin>", line 1, in <module> TypeError: only length-1 arrays can be converted to Python scalars How should I do this? -gideon

4 3

numpy save files from C
by John Hunter Feb. 26, 2009

Feb. 26, 2009

A colleague of mine has a bunch of numpy arrays saved with np.save and he now wants to access them directly in C, with or w/o the numpy C API doesn't matter. Does anyone have any sample code lying around which he can borrow from? The array is a structured array with an otherwise plain vanilla dtype (ints and floats). I've referred him to the npy-format NEP document, as well as the format.py implementation, so he can roll his own if need be, but if someone has a head start code example that would be great. Thanks, JDH

2 1

Fancy indexing question:
by Christopher Barker Feb. 26, 2009

Feb. 26, 2009

HI all, I'm having a bit of trouble getting fancy indexing to do what I want. Say I have a 2-d array: >>> a array([[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11], [12, 13, 14, 15], [16, 17, 18, 19], [20, 21, 22, 23]]) I want to extract a sub-array: The 1st, 3rd, and 4th rows: >>> i [1, 3, 4] and the 1st and 3rd columns: >>> j [1, 3] so I should get a 3x2 array: [[ 5, 7], [13, 15], [17, 19]] The obvious (to me!) way to do this: >>> a[i,j] Traceback (most recent call last): File "<stdin>", line 1, in <module> ValueError: shape mismatch: objects cannot be broadcast to a single shape fails. I can do: extract the rows: >>> rows = a[i] >>> rows array([[ 4, 5, 6, 7], [12, 13, 14, 15], [16, 17, 18, 19]]) then the columns: >>> sub_array = rows[:,j] >>> sub_array array([[ 5, 7], [13, 15], [17, 19]]) or, of course: >>> sub_array = a[i][:,j] >>> sub_array array([[ 5, 7], [13, 15], [17, 19]]) however, it seems I am doing one more data copy and array creation than I want -- can this be done as a single indexing operation? thanks, Chris -- Christopher Barker, Ph.D. Oceanographer Emergency Response Division NOAA/NOS/OR&R (206) 526-6959 voice 7600 Sand Point Way NE (206) 526-6329 fax Seattle, WA 98115 (206) 526-6317 main reception Chris.Barker(a)noaa.gov

4 5

What's wrong with floatint.c example code?
by Ralph Heinkel Feb. 26, 2009

Feb. 26, 2009

Hi, I'm trying to get into the realm of implementing my own numpy data types in numpy, and doing so I had a look at the floatint.c example coming from the numpy/doc/newdtype_example directory. Obviously it is not possible to create an array with the new floatint type by doing array([1,2,3,4,5,6,7,8], dtype=floatint) but instead this works: array([1,2,3,4,5,6,7,8]).view(ff.floatint_type) Would anybody mind helping me to understand this behavior? Is the reason for this that the floatint.c stuff is incomplete and would need some more elaboration in order to function as a proper type? Thanks in advance Ralph

1 0

CROSS-COMPILING NUMPY
by Yepez, Esteban Feb. 26, 2009

Feb. 26, 2009

Hello, I have having a hard time finding more information to help me, so I thought I'd ping you both. I need to x-compile numpy for a new processor. I have a toolchain, and have x-compiled python2.5.4, boost1.37,swig1.38. I just can't figure out how to x-compile numpy. Now I had to play many tricks to get python configured, so I may not have everything I need: For python, - I modified the Modules/Setup to build math and other modules that my target doesn't have - I modified configure so that it wouldn't crash when trying to run test programs - I modified the makefile so it would not try to build sharedmods Since Numpy uses python: python setup.py install, I figured that I could just use the host python to perform the build: >CC=my-gcc-compiler /usr/local/bin/python setup.py install Any help would be useful. Thanks Steve

3 4

coremath branch merged
by David Cournapeau Feb. 26, 2009

Feb. 26, 2009

Hi, I've just merged the work on the coremath branch into the trunk. Let me know if you see some problems - I did a quick sanity check on Linux after the merge, but I have not tested across many platforms, cheers, David

2 1