8d02cfe | Jasha Droppo | 08 June 2016, 00:33:24 UTC | Modify RNNNode to Interact Better with EvalDLL Interface | 08 June 2016, 00:33:24 UTC |
804b063 | Jasha Droppo | 07 June 2016, 16:06:15 UTC | Fix Merge Errors in Math*filter Project Files | 07 June 2016, 16:10:24 UTC |
7878a46 | Jasha Droppo | 06 June 2016, 18:52:48 UTC | Merge branch 'weixi/lfmmi' of https://github.com/Microsoft/CNTK into jdroppo/cudnn-rnn-lstm-lfmmi | 06 June 2016, 18:52:48 UTC |
7f7c1fc | Jasha Droppo | 27 May 2016, 15:43:54 UTC | Apply dataDecimationFactor to Cross Validaion by Shrinking Minibatch Not Decimating It | 27 May 2016, 15:43:54 UTC |
d743221 | Jasha Droppo | 27 May 2016, 00:56:52 UTC | Implement dataDecimationFactor In the process of placing the minibatch into the network, the minibatch first gets decimated by a rate of 1/dataDecimationFactor. This is useful when the input data is known to be highly correlated, for instance when training WBLSTM models for speech recognition. | 27 May 2016, 00:56:52 UTC |
09f7cd6 | Wayne Xiong | 07 April 2016, 00:32:49 UTC | LFMMI | 26 May 2016, 06:48:00 UTC |
c412b0d | Wayne Xiong | 24 May 2016, 21:33:23 UTC | Minor fixes in Convolution/BN | 26 May 2016, 06:47:54 UTC |
00d96ae | jeanfad | 25 May 2016, 14:33:56 UTC | fix div compatibility | 25 May 2016, 14:33:56 UTC |
ab07134 | jeanfad | 25 May 2016, 10:15:05 UTC | fixed failed unittests | 25 May 2016, 14:12:46 UTC |
e268730 | jeanfad | 25 May 2016, 10:11:18 UTC | fixed failed unittests | 25 May 2016, 14:12:45 UTC |
21d90a7 | jeanfad | 25 May 2016, 08:40:37 UTC | address cr comments | 25 May 2016, 14:12:44 UTC |
3b2f8ed | jeanfad | 24 May 2016, 13:47:29 UTC | fix is tensor test | 25 May 2016, 14:12:43 UTC |
9367d6a | jeanfad | 24 May 2016, 13:27:23 UTC | pass to identity rename | 25 May 2016, 14:12:42 UTC |
3509091 | jeanfad | 24 May 2016, 13:26:50 UTC | address CR comments | 25 May 2016, 14:12:41 UTC |
e2c61b1 | jeanfad | 24 May 2016, 09:56:56 UTC | clean up | 25 May 2016, 14:12:40 UTC |
672db34 | jeanfad | 24 May 2016, 09:29:47 UTC | add dropout to ops | 25 May 2016, 14:12:39 UTC |
001d858 | jeanfad | 24 May 2016, 08:14:19 UTC | add drop out to cntk2.py | 25 May 2016, 14:12:38 UTC |
bcb39e4 | jeanfad | 24 May 2016, 08:14:08 UTC | add dropout to cntk2 | 25 May 2016, 14:12:37 UTC |
ff19078 | jeanfad | 23 May 2016, 13:30:35 UTC | remove debuging print statements | 25 May 2016, 14:12:36 UTC |
5442b3d | jeanfad | 24 May 2016, 08:22:52 UTC | resolve conflicts after merging with slice change | 25 May 2016, 14:12:35 UTC |
15d21d1 | jeanfad | 20 May 2016, 17:37:37 UTC | add first version of transpose unit tests | 25 May 2016, 14:12:34 UTC |
deb6885 | jeanfad | 24 May 2016, 08:18:53 UTC | resolve conflicts with slice checkin | 25 May 2016, 14:12:33 UTC |
14be022 | jeanfad | 20 May 2016, 09:33:52 UTC | remove check broadcasting | 25 May 2016, 14:12:32 UTC |
b2b63e6 | jeanfad | 20 May 2016, 09:31:06 UTC | add docstring for overloaded ops | 25 May 2016, 14:12:31 UTC |
d3e3f7d | jeanfad | 20 May 2016, 09:22:12 UTC | address cr comments | 25 May 2016, 14:12:30 UTC |
b741c1a | jeanfad | 20 May 2016, 08:38:19 UTC | boadcast helper functions | 25 May 2016, 14:12:29 UTC |
6b76d16 | jeanfad | 19 May 2016, 17:08:30 UTC | helper functions for broadcasting test | 25 May 2016, 14:12:28 UTC |
b0fb9bd | jeanfad | 19 May 2016, 11:40:23 UTC | rename pass_tensor to identity | 25 May 2016, 14:12:27 UTC |
4694d40 | jeanfad | 19 May 2016, 10:02:33 UTC | add pass node | 25 May 2016, 14:12:26 UTC |
c10f51a | jeanfad | 18 May 2016, 11:53:58 UTC | add note to doc string of sqrt and log | 25 May 2016, 14:12:25 UTC |
6168668 | jeanfad | 18 May 2016, 11:37:02 UTC | remove power | 25 May 2016, 14:12:24 UTC |
7c4cd80 | jeanfad | 17 May 2016, 14:41:00 UTC | remove wrongly generated files | 25 May 2016, 14:12:23 UTC |
48fa413 | jeanfad | 13 May 2016, 15:57:28 UTC | forward test for power | 25 May 2016, 14:12:22 UTC |
66c5f98 | jeanfad | 13 May 2016, 15:20:08 UTC | unit test for square | 25 May 2016, 14:12:21 UTC |
7219bfc | jeanfad | 13 May 2016, 15:14:59 UTC | unit test for sqrt | 25 May 2016, 14:12:20 UTC |
3c69f92 | jeanfad | 13 May 2016, 15:03:49 UTC | add log unit test | 25 May 2016, 14:12:19 UTC |
2942fb8 | jeanfad | 13 May 2016, 14:41:01 UTC | add log, exp, power, sqrt, square to the facade | 25 May 2016, 14:12:18 UTC |
847d85c | jeanfad | 13 May 2016, 14:06:34 UTC | remove deprecated ops_test.py | 25 May 2016, 14:12:17 UTC |
e983434 | jeanfad | 13 May 2016, 13:21:36 UTC | fix square in cntk2 bs | 25 May 2016, 14:12:16 UTC |
7f39263 | jeanfad | 13 May 2016, 13:15:58 UTC | add log, power, sqrt, square and exp to core bs cntk2 | 25 May 2016, 14:12:15 UTC |
f0d1374 | jeanfad | 13 May 2016, 10:01:48 UTC | factor out constants in eval_test | 25 May 2016, 14:12:14 UTC |
1790616 | jeanfad | 13 May 2016, 09:59:10 UTC | initialize directory in deferred context | 25 May 2016, 14:12:13 UTC |
2a7e313 | jeanfad | 12 May 2016, 22:33:06 UTC | remove default context possibliity from reader.py | 25 May 2016, 14:12:12 UTC |
fdad4c5 | jeanfad | 12 May 2016, 22:29:04 UTC | use with get_newcontext() as ctx for cntk.eval | 25 May 2016, 14:12:11 UTC |
eab9bfd | jeanfad | 12 May 2016, 22:11:49 UTC | get rid of default context | 25 May 2016, 14:12:10 UTC |
b519e5d | Project Philly | 25 May 2016, 14:08:56 UTC | Integrate gaizna/eval_unittests into master | 25 May 2016, 14:08:56 UTC |
596e11a | Clemens Marschner | 25 May 2016, 11:44:19 UTC | minor glitches | 25 May 2016, 11:44:19 UTC |
ff083d5 | Clemens Marschner | 25 May 2016, 11:43:11 UTC | minor glitches | 25 May 2016, 11:43:11 UTC |
2abf395 | Clemens Marschner | 25 May 2016, 11:39:05 UTC | Minor glitches | 25 May 2016, 11:39:05 UTC |
35986ae | Clemens Marschner | 25 May 2016, 11:30:38 UTC | Extend checks and tests for extended eval interface Removed m_numberOfSamples, which is redundant Added more checks Fixed and extended test cases | 25 May 2016, 11:30:38 UTC |
1a6dce7 | Project Philly | 25 May 2016, 09:54:41 UTC | Integrate alrezni/longlivethereader into master | 25 May 2016, 09:54:41 UTC |
d8be484 | Alexey Reznichenko | 24 May 2016, 14:25:39 UTC | Replace UCIFast with CNTKTestFormat as the default reader | 25 May 2016, 09:11:38 UTC |
0928b9f | Wayne Xiong | 24 May 2016, 21:33:23 UTC | Minor fixes in Convolution/BN | 24 May 2016, 23:12:56 UTC |
ce364a1 | Project Philly | 24 May 2016, 14:13:56 UTC | Integrate thilow/CompareNode2 into master | 24 May 2016, 14:13:56 UTC |
8c6c0a1 | Mark Hillebrand | 24 May 2016, 12:27:31 UTC | Source/Readers/ExperimentalHTKMLFReader/: more logging | 24 May 2016, 12:30:19 UTC |
f2b13eb | Thilo Will | 24 May 2016, 12:25:11 UTC | Another attempt to fix mpic++ problem with compareNode | 24 May 2016, 12:25:11 UTC |
708dcb8 | Thilo Will | 24 May 2016, 12:05:37 UTC | Trying to fix a problem related to mpic++ | 24 May 2016, 12:05:37 UTC |
bd087c1 | Alexey Reznichenko | 24 May 2016, 11:54:25 UTC | Add deprecation notice | 24 May 2016, 11:57:03 UTC |
8eaf091 | Thilo Will | 24 May 2016, 09:52:02 UTC | ComparsionNode: removed type aliases as they don't work with mpic++. Readded lost file comparison_test.py | 24 May 2016, 09:52:02 UTC |
2a333f3 | Thilo Will | 24 May 2016, 09:01:03 UTC | Merged with master | 24 May 2016, 09:01:03 UTC |
44a1e87 | Alexey Reznichenko | 24 May 2016, 07:42:11 UTC | Revert "Raise an exception in UCIFast contstructor" This reverts commit 9ad127dacdc447490c02ed78a6a99283c68d811c. | 24 May 2016, 07:42:11 UTC |
9ad127d | Alexey Reznichenko | 24 May 2016, 07:40:42 UTC | Raise an exception in UCIFast contstructor | 24 May 2016, 07:40:42 UTC |
6cef7b0 | Project Philly | 23 May 2016, 19:55:53 UTC | Integrate wilrich/slice into master | 23 May 2016, 19:55:53 UTC |
bdb1271 | Willi Richert | 23 May 2016, 18:23:53 UTC | Improved exception handling | 23 May 2016, 18:23:53 UTC |
7376f25 | Thilo Will | 23 May 2016, 14:11:47 UTC | Removed two lines that where accidentially added with previous commit | 23 May 2016, 14:11:47 UTC |
c140a1e | Thilo Will | 23 May 2016, 14:06:40 UTC | Adding Compare nodes (Less, Greater, ...) | 23 May 2016, 14:06:40 UTC |
0833a26 | Jasha Droppo | 20 May 2016, 18:20:46 UTC | CuDNN RNN Remove Unnecessary GetLength() Member Function | 20 May 2016, 18:20:46 UTC |
5558144 | Jasha Droppo | 20 May 2016, 18:20:07 UTC | Merge remote-tracking branch 'origin/master' into jdroppo/cudnn-rnn-lstm Conflicts: Source/Math/Math.vcxproj Source/Math/Math.vcxproj.filters Source/Math/MathCUDA.vcxproj.filters | 20 May 2016, 18:20:07 UTC |
bd74679 | Eldar Akchurin | 20 May 2016, 15:12:23 UTC | Addressing CR | 20 May 2016, 15:12:23 UTC |
656b7bc | Eldar Akchurin | 20 May 2016, 14:22:28 UTC | Moving reindexing out of bundler into the cntk format deserializer | 20 May 2016, 14:22:28 UTC |
feb7ac0 | Gaizka Navarro | 20 May 2016, 07:20:48 UTC | Adds the EvalTests unit test project to the solution Adds tests for the EvalTests unittest project covering the new EvalExtended interface Adds check for mbLayout when creating VariableLayout in CNTKEvalExtended::ToVariableLayout function | 20 May 2016, 14:09:44 UTC |
4a3e8da | Willi Richert | 20 May 2016, 13:43:54 UTC | Normalize booleans in reader config | 20 May 2016, 13:56:16 UTC |
00a4c0c | Willi Richert | 20 May 2016, 13:42:14 UTC | Arg doc | 20 May 2016, 13:56:16 UTC |
317fa84 | Willi Richert | 20 May 2016, 13:41:41 UTC | Clean up examples | 20 May 2016, 13:56:15 UTC |
1f0ac9c | Willi Richert | 20 May 2016, 13:16:30 UTC | Updated auto-generated operator modules using the fixed fetch_ops | 20 May 2016, 13:56:14 UTC |
ef2a604 | Willi Richert | 20 May 2016, 13:16:01 UTC | Fix Trace() initialization | 20 May 2016, 13:56:13 UTC |
9e13e33 | Willi Richert | 20 May 2016, 13:15:05 UTC | Fix fetch_ops to work with bogus operator definitions | 20 May 2016, 13:56:13 UTC |
7cfc54c | Willi Richert | 18 May 2016, 12:43:27 UTC | Support for slice operator | 20 May 2016, 13:56:12 UTC |
4b5dd5f | Eldar Akchurin | 20 May 2016, 09:54:32 UTC | Adding more comments and tests | 20 May 2016, 11:51:38 UTC |
556fd99 | Eldar Akchurin | 20 May 2016, 08:45:14 UTC | Fixing some bugs | 20 May 2016, 11:51:38 UTC |
618e98a | Eldar Akchurin | 19 May 2016, 16:14:11 UTC | Adding support of sequence keys to CNTK key | 20 May 2016, 11:51:38 UTC |
07165dd | Eldar Akchurin | 19 May 2016, 12:36:04 UTC | Adding more data | 20 May 2016, 11:51:38 UTC |
f76a944 | Alexey Reznichenko | 20 May 2016, 08:03:29 UTC | Disable flaky test patterns | 20 May 2016, 10:11:16 UTC |
1360f0f | Jasha Droppo | 19 May 2016, 21:18:06 UTC | CuDNN RNN move to CUDNN5 Release Version cudnn-7.5-windows10-x64-v5.0-ga This required the following changes - CuDnnRNN::SetLength no longer necessary, the RNN descriptor is length-agnostic. - The sequence length is now a parameter of the backward and forward cudnn calls, so it is stored as a member function in the executor object. It is reset at each call of the forward pass. - dims[] and strides[] in cudnnSetTensorNDDescriptor have been re-arranged. But, data layout should be the same. - other API calls experienced minor changes | 19 May 2016, 21:18:06 UTC |
8dcfa43 | Jasha Droppo | 19 May 2016, 20:26:48 UTC | CuDNN RNN Allow Parameters To Have Non-Vector Shape I've noticed that 1-bit parallel training doesn't work well with this new RNN node. One possibility is that because the parameters are in a single column vector, they only get one "scale parameter" during the compressed exchange. This change allows the parameters to exist in any shape. Spreading them over more columns means more scale parameters, and hopefully better convergence. | 19 May 2016, 20:26:48 UTC |
b9f3c54 | Project Philly | 19 May 2016, 19:32:28 UTC | Integrate clemensm/s2sfix into master | 19 May 2016, 19:32:28 UTC |
2f2ca02 | Dong Yu | 19 May 2016, 17:50:58 UTC | Merge branch 'master' of https://github.com/Microsoft/CNTK into dongyu/rollbackfix | 19 May 2016, 17:50:58 UTC |
76c4bd8 | Alexey Reznichenko | 19 May 2016, 12:48:27 UTC | Add examples converted from UCI to the new text format Add/convert all examples from Examples/Other/Simple2d and several from Examples/Image (all that use UCIFastReader). | 19 May 2016, 16:32:11 UTC |
832655a | Clemens Marschner | 19 May 2016, 15:41:17 UTC | Review comments | 19 May 2016, 15:41:17 UTC |
2d9d1fa | Eldar Akchurin | 19 May 2016, 10:18:14 UTC | Addressing CR | 19 May 2016, 10:18:14 UTC |
80e8d73 | Eldar Akchurin | 19 May 2016, 09:28:23 UTC | Adding writer test | 19 May 2016, 09:28:23 UTC |
205c03f | Jasha Droppo | 18 May 2016, 22:39:44 UTC | CuDNN RNN Fix Typo | 18 May 2016, 22:39:44 UTC |
847d103 | Jasha Droppo | 18 May 2016, 22:37:08 UTC | CuDNN RNN Add Explicit Instantiaion of CuDnnRNN() Class | 18 May 2016, 22:37:08 UTC |
464445b | Jasha Droppo | 18 May 2016, 22:12:17 UTC | CuDNN RNN - Unified WLSTM and LSTM Functionality. - Clean up the Math API Calls to remove un-necessary details - Automatically use WLSTM (frame mode) when GetNumTimeSteps()==1 - Factored common code into CuDnnRNNExecutor::SetDescriptors() - CuDnnRNN::SetLength now does nothing if the length hasn't changed. - Input/Output shapes are determinied by: input feature length, output feature length, and a vector describing per-frame minibatch length. | 18 May 2016, 22:12:17 UTC |
ae0eb49 | Mark Hillebrand | 11 May 2016, 09:07:20 UTC | Get Tests/EndToEndTests/Examples/Speech/TIMIT/Write*/ to work again Note: WriteScaledLogLike produces results of varying checksum on Windows, where our hardware varies greater. If we see similar effects on Linux or WriteBottleneck as well, we need to rework. | 18 May 2016, 19:20:01 UTC |
5dd6b5f | Project Philly | 18 May 2016, 19:17:37 UTC | Integrate mahilleb/TestDisable into master | 18 May 2016, 19:17:37 UTC |
512216d | Mark Hillebrand | 18 May 2016, 09:04:15 UTC | Disable / work-around flaky tests pending further investigation. | 18 May 2016, 18:26:07 UTC |
7b6742a | Jasha Droppo | 18 May 2016, 17:20:32 UTC | CuDNN RNN Fix String Format Bug in InvalidArgument() Usage | 18 May 2016, 17:20:32 UTC |
83c152f | Clemens Marschner | 18 May 2016, 16:27:35 UTC | Documentation Fixes | 18 May 2016, 16:27:35 UTC |
402fee1 | Clemens Marschner | 18 May 2016, 15:22:11 UTC | Ignore some python artifacts | 18 May 2016, 15:55:10 UTC |