[incubator-mxnet] branch master updated: [MXNET-258] Improve flaky test_random.test_shuffle (#10313)

[zh...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

zhasheng pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git


The following commit(s) were added to refs/heads/master by this push:
     new 201ec00  [MXNET-258] Improve flaky test_random.test_shuffle (#10313)
201ec00 is described below

commit 201ec0063c29b95648e7b1e5dffdeb2eec82d050
Author: Deokjae Lee <36...@users.noreply.github.com>
AuthorDate: Sat Mar 31 14:40:26 2018 +0900

    [MXNET-258] Improve flaky test_random.test_shuffle (#10313)
    
    * Fix random seed for test_shuffle
    
    * A Comment for the fixed seed is added
    
    * Set random seed to reproduce the failure of test_random.test_shuffle
    
    * Add a link to the github issue on the discussion for the fixed seed
    
    * Testing with the failed random seed and increased number of samples
    
    * Increase the number of samples instead of fixing the random seed for test_shuffle
    
    * Increase the number of samples instead of fixing the random seed for test_shuffle
---
 tests/python/unittest/test_random.py | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/tests/python/unittest/test_random.py b/tests/python/unittest/test_random.py
index c8dc3c9..5138728 100644
--- a/tests/python/unittest/test_random.py
+++ b/tests/python/unittest/test_random.py
@@ -554,6 +554,7 @@ def test_zipfian_generator():
     mx.test_utils.assert_almost_equal(exp_cnt_sampled.asnumpy(), exp_cnt[sampled_classes].asnumpy(), rtol=1e-1, atol=1e-2)
     mx.test_utils.assert_almost_equal(exp_cnt_true.asnumpy(), exp_cnt[true_classes].asnumpy(), rtol=1e-1, atol=1e-2)
 
+# Issue #10277 (https://github.com/apache/incubator-mxnet/issues/10277) discusses this test.
 @with_seed()
 def test_shuffle():
     def check_first_axis_shuffle(arr):
@@ -596,7 +597,8 @@ def test_shuffle():
         # The outcomes must be uniformly distributed.
         # If `repeat2` is not large enough, this could fail with high probability.
         for p in itertools.permutations(range(0, data.size - stride + 1, stride)):
-            assert abs(1. * count[str(mx.nd.array(p))] / repeat2 - 1. / math.factorial(data.shape[0])) < 0.01
+            err = abs(1. * count[str(mx.nd.array(p))] / repeat2 - 1. / math.factorial(data.shape[0]))
+            assert err < 0.01, "The absolute error {} is larger than the tolerance.".format(err)
         # Check symbol interface
         a = mx.sym.Variable('a')
         b = mx.sym.random.shuffle(a)
@@ -622,9 +624,9 @@ def test_shuffle():
         assert len(count) == repeat
 
     # Test small arrays with different shapes
-    testSmall(mx.nd.arange(0, 3), 100, 20000)
-    testSmall(mx.nd.arange(0, 9).reshape((3, 3)), 100, 20000)
-    testSmall(mx.nd.arange(0, 18).reshape((3, 2, 3)), 100, 20000)
+    testSmall(mx.nd.arange(0, 3), 100, 40000)
+    testSmall(mx.nd.arange(0, 9).reshape((3, 3)), 100, 40000)
+    testSmall(mx.nd.arange(0, 18).reshape((3, 2, 3)), 100, 40000)
     # Test larger arrays
     testLarge(mx.nd.arange(0, 100000).reshape((10, 10000)), 10)
     testLarge(mx.nd.arange(0, 100000).reshape((10000, 10)), 10)

-- 
To stop receiving notification emails like this one, please contact
zhasheng@apache.org.