Mailman 3 February 2009 - NumPy-Discussion

Build with MKL on linux
by Neal Becker Feb. 20, 2009

Feb. 20, 2009

Trying to build numpy-1.2.1 with intel mkl 10.1.1.019 on linux F10 x86_64. echo $LD_LIBRARY_PATH /opt/intel/mkl/10.1.1.019/lib/em64t strace -e trace=file python -c 'import numpy; numpy.test()' 2>stuff Running unit tests for numpy NumPy version 1.2.1 NumPy is installed in /usr/lib64/python2.5/site-packages/numpy Python version 2.5.2 (r252:60911, Sep 30 2008, 15:42:03) [GCC 4.3.2 20080917 (Red Hat 4.3.2-4)] nose version 0.10.3 MKL FATAL ERROR: Cannot load neither libmkl_mc.so nor libmkl_def.so strace shows: ... open("/usr/lib64/python2.5/site-packages/numpy/core/tests/test_blasdot.py", O_RDONLY) = 3 open("/usr/lib64/python2.5/site-packages/numpy/core/tests/test_blasdot.pyc", O_RDONLY) = 5 open("/opt/intel/mkl/10.1.1.019/lib/em64t/libmkl_mc.so", O_RDONLY) = 3 open("/opt/intel/mkl/10.1.1.019/lib/em64t/libmkl_mc.so", O_RDONLY) = 3 open("/opt/intel/mkl/10.1.1.019/lib/em64t/libmkl_def.so", O_RDONLY) = 3 open("/opt/intel/mkl/10.1.1.019/lib/em64t/libmkl_def.so", O_RDONLY) = 3 The files are there and found, what could be wrong?

2 1

Update webpage for python requirements for Numpy/SciPy
by Bruce Southey Feb. 20, 2009

Feb. 20, 2009

Hi, Could someone please update the website to clearly state that numpy 1.2 requires Python 2.4 or later? I know it is in the release notes but that assumes people read them :-) It would be great to state this on the download and installation pages: http://www.scipy.org/Download http://www.scipy.org/Installing_SciPy Also there should be a mention that nose is required at for testing. A couple of dated material that I noticed are: http://www.scipy.org/Installing_SciPy/BuildingGeneral 'To build SciPy, Python version 2.3 or newer is required.' The FAQ page (http://www.scipy.org/FAQ) NumPy/SciPy installation 'Prerequisities NumPy requires the following software installed: 1. Python 2.3.x or 2.4.x or 2.5.x ' Thanks Bruce

4 4

Trac is overloaded
by David Cournapeau Feb. 20, 2009

Feb. 20, 2009

Hi, Just to mention that It looks like numpy trac is overloaded (error 500 and locked database) cheers, David

2 1

Initialize numpy array with other numpy arrays
by Frank Peacock Feb. 20, 2009

Feb. 20, 2009

Many thanks Robert. Frank

1 0

numpy 1.2.1 uploaded to unstable
by Ondrej Certik Feb. 19, 2009

Feb. 19, 2009

Hi, I finally found time to upload the numpy 1.2.1 to Debian unstable (currently it's in incoming: http://incoming.debian.org/). The package is lintian clean, but there is one test that failed for me in chroot. I'll wait until it gets to mirrors and then try it on my laptop and report a bug (I uploaded from a ubuntu machine, but of course I compiled it in pbuilder with sid and tested in chroot). Ondrej

1 0

parallel compilation of numpy
by Ondrej Certik Feb. 19, 2009

Feb. 19, 2009

Hi, I have a shiny new computer with 8 cores and numpy still takes forever to compile --- is there a way to compile it in parallel (make -j9)? Do distutils allow that? If not, let's move to some build system that allows that? Just wanted to check if there is some reason for that, apart from patches welcome. :) Ondrej

12 24

linalg.norm along axis?
by Grissiom Feb. 19, 2009

Feb. 19, 2009

Hi all, Is there any possibility to calculate norm along axis? For example: a = np.array(( (3,4), (6,8))) And I want to get: array([5.0, 10.0]) I currently use a for loop to achieve this, Is there any more elegant way to do this? -- Cheers, Grissiom

2 2

rgb_to_hsv in scipy.misc ? (was: Optimizing speed for large-array inter-element algorithms (specifically, color space conversion))
by Nicolas Pinto Feb. 18, 2009

Feb. 18, 2009

Hello, Would it be possible to include the following rgb to hsv conversion code in scipy (probably in misc along with misc.imread, etc.) ? What do you think? Thanks in advance. Best regards, -- Nicolas Pinto Ph.D. Candidate, Brain & Computer Sciences Massachusetts Institute of Technology, USA http://web.mit.edu/pinto # ------------------------------------------------------------------------------ import numpy as np def rgb_to_hsv_arr(arr): """ fast rgb_to_hsv using numpy array """ # adapted from Arnar Flatberg # http://www.mail-archive.com/numpy-discussion@scipy.org/msg06147.html # it now handles NaN properly and mimics colorsys.rgb_to_hsv output arr = arr/255. out = np.empty_like(arr) arr_max = arr.max(-1) delta = arr.ptp(-1) s = delta / arr_max s[delta==0] = 0 # red is max idx = (arr[:,:,0] == arr_max) out[idx, 0] = (arr[idx, 1] - arr[idx, 2]) / delta[idx] # green is max idx = (arr[:,:,1] == arr_max) out[idx, 0] = 2. + (arr[idx, 2] - arr[idx, 0] ) / delta[idx] # blue is max idx = (arr[:,:,2] == arr_max) out[idx, 0] = 4. + (arr[idx, 0] - arr[idx, 1] ) / delta[idx] out[:,:,0] = (out[:,:,0]/6.0) % 1.0 out[:,:,1] = s out[:,:,2] = arr_max # rescale back to [0, 255] out *= 255. # remove NaN out[np.isnan(out)] = 0 return out

1 0

additional dtype argument to numpy.dot() (Re: Numpy-discussion Digest, Vol 29, Issue 48)
by David Henderson Feb. 18, 2009

Feb. 18, 2009

Hi Paul, list: Thanks for the reply. numpy.sum() does indeed have a dtype for the accumulator for the sum. numpy.sum() does not implement an inner (dot) product, just a straight summation. The feature I'm requesting is to add a similar accumulator type argument for numpy.dot(). After tracing some of the code, the there is a generic dot product algorithm for the various types. If a BLAS library is available, the BLAS routines will supplant the library, but only for the BLAS routines that exist. BLAS doesn't support long double, so only a subset of the different calls for different types call BLAS. The place where this feature comes into play is iterative refinement of a linear algebra solution. To implement iterative refinement, an error term is calculated as a matrix dot product with a higher precision than the elements of the matrix. The floating point hardware on the i386/PPC/Sparc is ready to go for this application, its just that BLAS doesn't implement it. The numpy.dot() function is missing the feature to supply the type of the accumulator for the sum. I'm looking for a way to prototype some routines to perform iterative refinement, and python/numpy looks like an ideal platform. numpy has _almost_ the feature set needed to implement this. Adding the accumulator type argument to numpy.dot() is a needed feature. code snippet exhibiting float128 functionality that already exists: -------------------------- Python 2.6.1 (r261:67515, Feb 12 2009, 16:56:09) [GCC 4.0.1 (Apple Computer, Inc. build 5367)] on darwin Type "help", "copyright", "credits" or "license" for more information. >>> import numpy >>> z=numpy.longdouble((1,2,3)) >>> z array([1.0, 2.0, 3.0], dtype=float128) >>> y=numpy.double((4,5,6)) >>> y array([ 4., 5., 6.]) >>> numpy.dot(y,y) 77.0 >>> numpy.dot(y,z) 32.0 >>> numpy.dot(z,z) 14.0 --------------------------- I'm prepared to make the changes to numpy. It looks like they would entail modifying the source file "multiarrayobject.c". The existing code is: /*NUMPY_API Numeric.innerproduct(a,v) */ static PyObject * PyArray_InnerProduct(PyObject *op1, PyObject *op2) { If I make changes, can I get them inserted into the numpy source base (after appropriate testing??) Thanks in advance, David Henderson On Feb 14, 2009, at 8:04 AM, numpy-discussion-request(a)scipy.org wrote: > Fri, 13 Feb 2009 12:04:12 -0800, David Henderson wrote: >> I'd like accumulate the summation in extended precision, "double" sum >> for float inputs, "long double" sum for "double" inputs. >> >> A way of doing this is to add an optional third argument to dot - to >> specify the summation type. > > `dot` does matrix-matrix, matrix-vector, or vector-vector products. > These > are usually implemented via calling the BLAS linear algebra > library, and > AFAIK BLAS only has routines for type-homogenous arguments. So I doubt > this can be implemented for `dot` in any way. > > On the other hand, `numpy.sum` already has a `dtype` argument that > specifies the accumulator data type. Maybe you can use that? > > -- > Pauli Virtanen > ----------------------------- >

2 1

Ironclad v0.8.1 released
by William Reade Feb. 18, 2009

Feb. 18, 2009

Hi all I'm fairly pleased to announce the release of Ironclad v0.8.1; it's not an enormous technical leap above v0.8, but it does now enable you to import and use SciPy and Matplotlib with IronPython on Win32 (with some restrictions; see project page) . Downloads, and more details, are available at http://code.google.com/p/ironclad -- please do play with it, and let me know if you have any problems. Cheers William

1 0