Mailman 3 January 2011 - SciPy-Dev

Re: [SciPy-Dev] postponing 0.9.0rc1 by one week
by Fabian Pedregosa 17 Jan '11

17 Jan '11

On Sun, Jan 16, 2011 at 3:59 PM, Ralf Gommers <ralf.gommers(a)googlemail.com> wrote: > Hi all, > > There is still one release blocker for 0.9.0, the change in behavior of > convolve/correlate. See the old_behavior keyword in > correlate/convolve/correlate2d/convolve2d, and also > http://projects.scipy.org/scipy/ticket/1301. Therefore the release of RC1 is > postponed by one week. If anyone has time to look at this, that would be > great. > > Three other things that could be fixed as well are the following: > 1. Medfilt test failure on linux 64-bit, see > http://news.gmane.org/gmane.comp.python.scientific.devel. This will be > marked as knownfail for this release if no one has time to investigate. > 2. CorrelateComplex192 failures on Windows, see the other email I just sent. > 3. Test warnings from interpolate and linalg, see below. Will just be > filtered out for 0.9.0 unless they are fixed in time. > > /Users/rgommers/Code/scipy/scipy/interpolate/fitpack2.py:673: UserWarning: > The coefficients of the spline returned have been computed as the > minimal norm least-squares solution of a (numerically) rank deficient > system (deficiency=7). If deficiency is large, the results may be > inaccurate. Deficiency may strongly depend on the value of eps. > warnings.warn(message) > ....../Users/rgommers/Code/scipy/scipy/interpolate/fitpack2.py:604: > UserWarning: > The required storage space exceeds the available storage space: nxest > or nyest too small, or s too small. > The weighted least-squares spline corresponds to the current set of > knots. > warnings.warn(message) > > /Users/rgommers/Code/scipy/scipy/linalg/decomp_qr.py:91: ComplexWarning: > Casting complex values to real discards the imaginary part > qr, tau, work, info = geqrf(a1, lwork=lwork, overwrite_a=overwrite_a) > /Users/rgommers/Code/scipy/scipy/linalg/decomp_qr.py:126: ComplexWarning: > Casting complex values to real discards the imaginary part > Q, work, info = gor_un_gqr(qqr, tau, lwork=lwork, overwrite_a=1) These warnings come from the fact that lwork returned from a workspace query has the same dtype as the principal matrix, but when we call the computational routine it expects an integer value. A simple workaround is here: https://github.com/fabianp/scipy-svn/commit/5942b71bd321b0960fd179c9785273c… Although it would be more correct to convert it also to integer, i.e., lwork = int(work[0].real) Fabian.

1 0

[PATCH] improvements for rq factorization
by Fabian Pedregosa 17 Jan '11

17 Jan '11

Hi all. In this patch I implement several improvements for the RQ matrix factorization in function linalg.rq: - API: now uses the same keywords as qr function: implemented mode='economic' - type support: now supports non-square as well as complex arrays. This way I could uncomment a lot of tests. - performance: now uses LAPACK orgrq for computing the Q matrix instead of plain matrix multiplications. On a 500 x 500 array in my Core Duo, this makes it go from 17.6s s to 0.16s, i.e. gain of a 100x factor. - Also, some docstring fixes in linalg.qr function. Best, Fabian.

1 1

unicode in THANKS.txt
by josef.pktd＠gmail.com 15 Jan '11

15 Jan '11

What's the policy for unicode in text files? I added some contributers to THANKS.txt, but have problems with an accented name, Ernest Adrogué Is it possible to "internationalize" THANKS.txt ? Josef

2 2

backporting scipy.stats fixes
by josef.pktd＠gmail.com 15 Jan '11

15 Jan '11

I added 3 changesets in trunk, that should be backported to 0.9 7030 is a trivial fix to avoid an exception 7028 and 7029 correct the integration bounds in the expect methods and have, I think, sufficient test cases. Thanks, Josef

2 2

updating numpy scipy crash - numpy on pypi
by josef.pktd＠gmail.com 15 Jan '11

15 Jan '11

I *finally* updated numpy and scipy. For numpy 1.5.1, I used the files on pypi. When I installed scipy 0.9b1 it crashed (segfaulted) in the testsuite. With the superpack installer for numpy from the sourceforge download, scipy 0.9b1 passes all tests Ran 4806 tests in 287.843s OK (KNOWNFAIL=12, SKIP=35) <nose.result.TextTestResult run=4806 errors=0 failures=0> It looks like the non-superpack numpy files on pypi are not compatible with scipy in my setup (win32, sse2, python 2.5.2) Another question, are the clapack tests skipped on purpose, for example test_clapack_dsygv_1 (test_gesv.TestSygv) ... SKIP: Skipping test: test_clapack_ dsygv_1 Clapack empty, skip flapack test test_clapack_dsygv_2 (test_gesv.TestSygv) ... SKIP: Skipping test: test_clapack_ dsygv_2 Clapack empty, skip flapack test test_clapack_dsygv_3 (test_gesv.TestSygv) ... SKIP: Skipping test: test_clapack_ dsygv_3 Clapack empty, skip flapack test test_clapack_ssygv_1 (test_gesv.TestSygv) ... SKIP: Skipping test: test_clapack_ ssygv_1 Clapack empty, skip flapack test test_clapack_ssygv_2 (test_gesv.TestSygv) ... SKIP: Skipping test: test_clapack_ ssygv_2 Clapack empty, skip flapack test test_clapack_ssygv_3 (test_gesv.TestSygv) ... SKIP: Skipping test: test_clapack_ ssygv_3 Clapack empty, skip flapack test Josef

2 5

kde evaluate and dtypes, atleast_float ?
by josef.pktd＠gmail.com 15 Jan '11

15 Jan '11

http://projects.scipy.org/scipy/ticket/1181 the dtypes in kde evaluate only makes sense if we have at least float integer input produces nonsense evaluate has points = atleast_2d(points).astype(self.dataset.dtype) .... result = zeros((m,), points.dtype) .... result += exp(-energy) I don't know which dtypes kde is supposed to handle, but what's the best way to guarantee that the result is "atleast_float", so that a probability makes sense. I don't know what a complex probability is, but handling complex is useful for complex derivatives. Thanks, Josef

1 0

one failure in scipy trunk
by josef.pktd＠gmail.com 14 Jan '11

14 Jan '11

compiled against numpy 1.5.1 Josef ====================================================================== FAIL: Test singular pair ---------------------------------------------------------------------- Traceback (most recent call last): File "c:\programs\python25\lib\site-packages\nose-0.11.1-py2.5.egg\nose\case.p y", line 183, in runTest self.test(*self.arg) File "C:\Josef\_progs\Subversion\scipy_trunk_g09\dist\scipy-0.10.0.dev7027.win 32\Programs\Python25\Lib\site-packages\scipy\linalg\tests\test_decomp.py", line 200, in test_singular self._check_gen_eig(A, B) File "C:\Josef\_progs\Subversion\scipy_trunk_g09\dist\scipy-0.10.0.dev7027.win 32\Programs\Python25\Lib\site-packages\scipy\linalg\tests\test_decomp.py", line 189, in _check_gen_eig err_msg=msg) File "C:\Programs\Python25\Lib\site-packages\numpy\testing\utils.py", line 774 , in assert_array_almost_equal header='Arrays are not almost equal') File "C:\Programs\Python25\Lib\site-packages\numpy\testing\utils.py", line 618 , in assert_array_compare raise AssertionError(msg) AssertionError: Arrays are not almost equal array([[22, 34, 31, 31, 17], [45, 45, 42, 19, 29], [39, 47, 49, 26, 34], [27, 31, 26, 21, 15], [38, 44, 44, 24, 30]]) array([[13, 26, 25, 17, 24], [31, 46, 40, 26, 37], [26, 40, 19, 25, 25], [16, 25, 27, 14, 23], [24, 35, 18, 21, 22]]) (mismatch 50.0%) x: array([ 3.57527348e-16 -2.30736377e-08j, 3.57527348e-16 +2.30736377e-08j, 2.86330799e-01 +0.00000000e+00j, 2.00000000e+00 +0.00000000e+00j]) y: array([ -1.31566605e-01+0.j, -3.23644701e-08+0.j, 3.23644694e-08+0.j, 2.00000000e+00+0.j]) ---------------------------------------------------------------------- Ran 4810 tests in 277.406s FAILED (KNOWNFAIL=12, SKIP=35, failures=1) <nose.result.TextTestResult run=4810 errors=0 failures=1>

3 6

speed of nosetests scipy.stats
by josef.pktd＠gmail.com 13 Jan '11

13 Jan '11

nosetests scipy.stats includes the "slow" tests, but if I remember correctly they ran in around 7 minutes on my computer. Now I get: Ran 1572 tests in 2479.594s Is it slow for anyone else, or is it just because my computer is busy with other things also. I gave up running my fit tests after about an hour or so. scipy 0.9b1, numpy 1.5.1 official installers and what kind of operation produces "Warning: invalid value encountered in subtract" ? I get hundreds of these Josef

6 9

review request - test noise clean up
by Ralf Gommers 13 Jan '11

13 Jan '11

Hi, I cleaned most of the noise in the test output (only stats module and a few random ones left): https://github.com/rgommers/scipy/tree/testnoise Can someone please check this and confirm that this can go into trunk as well as 0.9.x? Especially for the scipy.special commit. There are a few warnings that I'll only silence in 0.9.x because they indicate actual problems or should be fixed elsewhere (like astype() in numpy: http://projects.scipy.org/numpy/ticket/1709). For the ComplexWarning filter, is it preferable to be really granular with linenos like warnings.filterwarnings('ignore', category=np.ComplexWarning, lineno=98) or just filter it out of the whole module? Thanks, Ralf

6 9

Bug in KDTree
by Benjamin Root 11 Jan '11

11 Jan '11

Hello, I came across an issue using KDTree. I searched the bug list, and it appears that there has already been a patch submitted for review. I have added comments to the bug report and I think the patch is good to go. I am not an expert in KDTree myself, but I think the logic is sound. http://projects.scipy.org/scipy/ticket/1052 If the first node of the tree is a leafnode, then query_pairs() will fail if the points are closer than the distance threshold. This is because the algorithm is assuming that the first node will always be an innernode. By rearranging the if-statements so that a check for a leafnode is done first, you protect against such a situation. At least, that is my understanding. I welcome comments on this. Thanks, Ben Root

2 5