[GitHub] vandanavk commented on a change in pull request #12099: Fix precision issue of test case test_rnnrelu_bidirectional and test_rnnrelu_sym

[GitBox <gi...@apache.org>]

vandanavk commented on a change in pull request #12099: Fix precision issue of test case test_rnnrelu_bidirectional and test_rnnrelu_sym
URL: https://github.com/apache/incubator-mxnet/pull/12099#discussion_r208983616
 
 

 ##########
 File path: tests/python/unittest/test_operator.py
 ##########
 @@ -191,13 +191,12 @@ def test_rnnrelu_sym():
     stack.add(mx.rnn.RNNCell(H, activation='relu', prefix='l1_'))
     stack.add(mx.rnn.RNNCell(H, activation='relu', prefix='l2_'))
 
-    check_rnn_consistency(fused, stack, T, N, I, H, 'write')
-    check_rnn_consistency(fused, stack, T, N, I, H, 'add')
-    check_rnn_consistency(fused, stack, T, N, I, H, 'null')
-
+    check_rnn_consistency(fused, stack, T, N, I, H, 'write', rtol=1e-2, atol=1e-2)
+    check_rnn_consistency(fused, stack, T, N, I, H, 'add', rtol=1e-2, atol=1e-2)
+    check_rnn_consistency(fused, stack, T, N, I, H, 'null', rtol=1e-2, atol=1e-2)
 
 
 Review comment:
   Do we need to change the tolerance for test_rnnrelu_sym too? test_rnnrelu_sym passes for 10000 runs with the existing code (rtol=1e-2, atol=1e-4). Only test_rnnrelu_bidirectional fails with this tolerance.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services