Bug#669466: [NeuralEnsemble] PyNN: fresh tests failures with 0.7.2
OK, all the unit tests now pass on a fresh install of Debian Wheezy
(which has NumPy 1.6.2), as well as on Ubuntu 11.04 (NumPy 1.5.1).
I've made a new bug-fix release:
package: http://pypi.python.org/pypi/PyNN/0.7.4
source: https://neuralensemble.org/svn/PyNN/tags/0.7.4 or
https://neuralensemble.org/svn/PyNN/branches/0.7
Please let me know if there are any problems.
Cheers,
Andrew
On 2 juil. 12, at 23:38, Yaroslav Halchenko wrote:
oki doki -- here is an update -- boiled down all numpy-related
issues (I have
not looked at the failing mock-related test -- which smells also
like a
regression in mock 0.8.0):
1.
NB unittests.test_files.test_NumpyBinaryFile failure seems was gone
in
its own course with upgrades of numpy
but I also started to get
=
=
ERROR: unittests.test_files.test_NumpyBinaryFile
--
Traceback (most recent call last):
File /usr/lib/python2.7/dist-packages/nose/case.py, line 197, in
runTest
self.test(*self.arg)
File /tmp/buildd/pynn-0.7.2/test/unittests/test_files.py, line
92, in test_NumpyBinaryFile
assert_equal(nbf.get_metadata(), metadata)
File /tmp/buildd/pynn-0.7.2/build/lib.linux-x86_64-2.7/pyNN/
recording/files.py, line 216, in get_metadata
self.fileobj.seek(0)
ValueError: I/O operation on closed file
here is the report:
http://projects.scipy.org/numpy/ticket/2178#comment:1
and tentative fix
https://github.com/numpy/numpy/pull/328
pynn workaround would probably simply to disable that portions
of the test on numpy = 1.6, and = (where gets fixed).
agree?
2.
A new one
==
FAIL: unittests.test_files.test_StandardTextFile_write
--
Traceback (most recent call last):
File /usr/lib/python2.7/dist-packages/nose/case.py, line 197, in
runTest
self.test(*self.arg)
File /home/yoh/deb/gits/pkg-exppsy/pynn/test/unittests/
test_files.py, line 57, in test_StandardTextFile_write
target)
AssertionError: [call('# a = 1\n# b = 9.99\n'),
call('0.0\t2.2998\n'),
call('1.0\t3.3999\n'),
call('2.0\t4.2998\n')] != [(('# a = 1\n# b = 9.99\n',),
{}), (('0.0\t2.3\n',), {}), (('1.0\t3.4\n',), {}), (('2.0\t4.3\n',),
{})]
which I haven't figured out 100% but
$ python2.7 -c print repr([(0, 2.3),(1, 3.4),(2, 4.3)])
[(0, 2.3), (1, 3.4), (2, 4.3)]
$ python2.6 -c print repr([(0, 2.3),(1, 3.4),(2, 4.3)])
[(0, 2.2998), (1, 3.3999), (2,
4.2998)]
suggests that it might be just worth not relying on repr of FPs where
either decimal or binary form aren't unambiguous, so please
find/accept/consider attached patch or advise on the alternative
resolution
3.
=
=
FAIL: unittests.test_files.test_NumpyBinaryFile
--
Traceback (most recent call last):
File /usr/lib/python2.7/dist-packages/nose/case.py, line 197, in
runTest
self.test(*self.arg)
File /«PKGBUILDDIR»/test/unittests/test_files.py, line 92, in
test_NumpyBinaryFile
assert_equal(nbf.get_metadata(), metadata)
AssertionError: {'a': 1, 'b': 9} != {'a': 1, 'b': 9.99}
- {'a': 1, 'b': 9}
+ {'a': 1, 'b': 9.99}
? +++
boiled down once again to numpy's regression (fixed upstream in as
bisected
e15d0bdd3cc0bc0928e1f4d0b419a2fb3de02af9)
Debian bug report: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=679948
pynn resolution:
is metadata always of 'float' type? then dtype=(str, float) could be
provided to np.array within NumpyBinaryFile.save to assure correct
dtype...
please advise
--
Yaroslav O. Halchenko
Postdoctoral Fellow, Department of Psychological and Brain Sciences
Dartmouth College, 419 Moore Hall, Hinman Box 6207, Hanover, NH 03755
Phone: +1 (603) 646-9834 Fax: +1 (603) 646-1419
WWW: http://www.linkedin.com/in/yarik
--
You received this message because you are subscribed to the Google
Groups Neural Ensemble group.
To post to this group, send an email to neuralensem...@googlegroups.com
.
To unsubscribe from this group, send email to neuralensemble+unsubscr...@googlegroups.com
.
For more options, visit this group at http://groups.google.com/group/neuralensemble?hl=en-GB
.
0001-BF-test_StandardTextFile_write-test-on-numbers-with-.patch
--
Dr Andrew Davison
Unité de Neuroscience, Information et Complexité (UNIC)
Institut de Neurobiologie Alfred Fessard
Centre Nationale de la Recherche Scientifique
1, avenue de la Terrasse
91198 Gif sur Yvette
France
Tel: +33 1 69 82 34 51
http://www.andrewdavison.info/
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#669466: [NeuralEnsemble] PyNN: fresh tests failures with 0.7.2
Thank you Andrew! I have uploaded 0.7.4-1 to unstable -- now the job would be to persuade release team to allow it to enter wheezy ;-) On Tue, 03 Jul 2012, Andrew Davison wrote: OK, all the unit tests now pass on a fresh install of Debian Wheezy (which has NumPy 1.6.2), as well as on Ubuntu 11.04 (NumPy 1.5.1). I've made a new bug-fix release: package: http://pypi.python.org/pypi/PyNN/0.7.4 source: https://neuralensemble.org/svn/PyNN/tags/0.7.4 or https://neuralensemble.org/svn/PyNN/branches/0.7 Please let me know if there are any problems. Cheers, Andrew -- Yaroslav O. Halchenko Postdoctoral Fellow, Department of Psychological and Brain Sciences Dartmouth College, 419 Moore Hall, Hinman Box 6207, Hanover, NH 03755 Phone: +1 (603) 646-9834 Fax: +1 (603) 646-1419 WWW: http://www.linkedin.com/in/yarik -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
Bug#669466: [NeuralEnsemble] PyNN: fresh tests failures with 0.7.2
oki doki -- here is an update -- boiled down all numpy-related issues (I have
not looked at the failing mock-related test -- which smells also like a
regression in mock 0.8.0):
1.
NB unittests.test_files.test_NumpyBinaryFile failure seems was gone in
its own course with upgrades of numpy
but I also started to get
==
ERROR: unittests.test_files.test_NumpyBinaryFile
--
Traceback (most recent call last):
File /usr/lib/python2.7/dist-packages/nose/case.py, line 197, in runTest
self.test(*self.arg)
File /tmp/buildd/pynn-0.7.2/test/unittests/test_files.py, line 92, in
test_NumpyBinaryFile
assert_equal(nbf.get_metadata(), metadata)
File
/tmp/buildd/pynn-0.7.2/build/lib.linux-x86_64-2.7/pyNN/recording/files.py,
line 216, in get_metadata
self.fileobj.seek(0)
ValueError: I/O operation on closed file
here is the report:
http://projects.scipy.org/numpy/ticket/2178#comment:1
and tentative fix
https://github.com/numpy/numpy/pull/328
pynn workaround would probably simply to disable that portions
of the test on numpy = 1.6, and = (where gets fixed).
agree?
2.
A new one
==
FAIL: unittests.test_files.test_StandardTextFile_write
--
Traceback (most recent call last):
File /usr/lib/python2.7/dist-packages/nose/case.py, line 197, in runTest
self.test(*self.arg)
File /home/yoh/deb/gits/pkg-exppsy/pynn/test/unittests/test_files.py, line
57, in test_StandardTextFile_write
target)
AssertionError: [call('# a = 1\n# b = 9.99\n'),
call('0.0\t2.2998\n'),
call('1.0\t3.3999\n'),
call('2.0\t4.2998\n')] != [(('# a = 1\n# b = 9.99\n',), {}),
(('0.0\t2.3\n',), {}), (('1.0\t3.4\n',), {}), (('2.0\t4.3\n',), {})]
which I haven't figured out 100% but
$ python2.7 -c print repr([(0, 2.3),(1, 3.4),(2, 4.3)])
[(0, 2.3), (1, 3.4), (2, 4.3)]
$ python2.6 -c print repr([(0, 2.3),(1, 3.4),(2, 4.3)])
[(0, 2.2998), (1, 3.3999), (2, 4.2998)]
suggests that it might be just worth not relying on repr of FPs where
either decimal or binary form aren't unambiguous, so please
find/accept/consider attached patch or advise on the alternative resolution
3.
==
FAIL: unittests.test_files.test_NumpyBinaryFile
--
Traceback (most recent call last):
File /usr/lib/python2.7/dist-packages/nose/case.py, line 197, in runTest
self.test(*self.arg)
File /ëPKGBUILDDIRû/test/unittests/test_files.py, line 92, in
test_NumpyBinaryFile
assert_equal(nbf.get_metadata(), metadata)
AssertionError: {'a': 1, 'b': 9} != {'a': 1, 'b': 9.99}
- {'a': 1, 'b': 9}
+ {'a': 1, 'b': 9.99}
? +++
boiled down once again to numpy's regression (fixed upstream in as bisected
e15d0bdd3cc0bc0928e1f4d0b419a2fb3de02af9)
Debian bug report: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=679948
pynn resolution:
is metadata always of 'float' type? then dtype=(str, float) could be
provided to np.array within NumpyBinaryFile.save to assure correct dtype...
please advise
--
Yaroslav O. Halchenko
Postdoctoral Fellow, Department of Psychological and Brain Sciences
Dartmouth College, 419 Moore Hall, Hinman Box 6207, Hanover, NH 03755
Phone: +1 (603) 646-9834 Fax: +1 (603) 646-1419
WWW: http://www.linkedin.com/in/yarik
From 70295461024cc95709ae214b96aa4caa9c225ae9 Mon Sep 17 00:00:00 2001
From: Yaroslav Halchenko deb...@onerussian.com
Date: Mon, 2 Jul 2012 11:42:11 -0400
Subject: [PATCH] BF: test_StandardTextFile_write - test on numbers with a
disambigous binary/decimal repr
---
test/unittests/test_files.py | 17 +
1 file changed, 13 insertions(+), 4 deletions(-)
diff --git a/test/unittests/test_files.py b/test/unittests/test_files.py
index da4ebe8..1c937ce 100644
--- a/test/unittests/test_files.py
+++ b/test/unittests/test_files.py
@@ -46,12 +46,21 @@ def test_close():
def test_StandardTextFile_write():
files.open = Mock()
stf = files.StandardTextFile(filename, w)
-data=[(0, 2.3),(1, 3.4),(2, 4.3)]
+# yoh: %r on FP is inconsistent even across Python's versions:
+# python-2.7 -c print repr([(0, 2.3),(1, 3.4),(2, 4.3)])
+# [(0, 2.3), (1, 3.4), (2, 4.3)]
+# python2.6 -c print repr([(0, 2.3),(1, 3.4),(2, 4.3)])
+# [(0, 2.2998), (1, 3.3999), (2, 4.2998)]
+# so that makes straightforward comparison with numbers without
+# exact binary representation weak and leading to spurious
+# failures. So testing here basic functioning on numbers
