I agree that failing a build for a timing test such as this is
inherently unreliable...so let's turn it off.
(Tested only in sparse/tests/test_basic.py, not a full build.)
It may be that theano is slower because it's using a
lowest-common-denominator BLAS while numpy does runtime processor
detection, as recently discussed at
https://lists.debian.org/debian-science/2017/02/msg00043.html , but now
is not the time to be fixing that.
Description: Disable overly environment-dependent test
Testing speed by wall-clock time is inherently unreliable on a
shared machine such as Debian's buildds:
don't let it fail the whole build.
Author: Rebecca N. Palmer <rebecca_pal...@zoho.com>
Bug-Debian: https://bugs.debian.org/835531
Forwarded: not-needed
diff --git a/theano/sparse/tests/test_basic.py b/theano/sparse/tests/test_basic.py
index 8c183b9..03d79f1 100644
--- a/theano/sparse/tests/test_basic.py
+++ b/theano/sparse/tests/test_basic.py
@@ -1209,8 +1209,8 @@ class test_structureddot(unittest.TestCase):
overhead_tol = 0.002 # seconds
overhead_rtol = 1.1 # times as long
utt.assert_allclose(scipy_result, theano_result)
- if (not theano.config.mode in ["DebugMode", "DEBUG_MODE"] and
- theano.config.cxx):
+
+ if 0:#(not theano.config.mode in ["DebugMode", "DEBUG_MODE"] and theano.config.cxx):
self.assertFalse(
theano_time > overhead_rtol * scipy_time + overhead_tol,
(theano_time,