https://github.com/Microsoft/CNTK

sort by:
Revision Author Date Message Commit Date
d3db246 Merge branch 'master' into alrezni/text_reader_benchmark Conflicts: CNTK.sln 13 May 2016, 13:36:50 UTC
30d4a21 Work in progress 13 May 2016, 13:31:55 UTC
fb98b85 Integrate alrezni/scratch into master 13 May 2016, 12:34:00 UTC
bd4ccab Integrate mahilleb/ExpHtkmlfEnforceIdenticalBaselines into master 13 May 2016, 10:10:17 UTC
7a83e7e Tests/EndToEndTests/Speech/ExperimentalHtkmlfReader/: re-instate test for identical baselines (Reverts parts of 04abf66d78420c410da9bf9e4ba82d88223f95b1) 13 May 2016, 07:55:10 UTC
3600ed6 Add missing #define __STDC_FORMAT_MACROS Close #464 13 May 2016, 07:37:56 UTC
5f2cfa0 msra::asr::htkfeatreader::parsedpath: accept trailing carriage return 13 May 2016, 07:24:03 UTC
b67b157 Integrate thhoens/matfix into master 12 May 2016, 18:23:33 UTC
a99c334 guard against faulty input data 12 May 2016, 09:35:58 UTC
94e8115 address CR comments 12 May 2016, 09:35:56 UTC
fb91d07 remove no longer needed template file for global params 12 May 2016, 09:35:55 UTC
fcc6663 fix bug in config generation for sgd parallele training 12 May 2016, 09:35:54 UTC
98a6b08 add doc string for SGD 12 May 2016, 09:35:53 UTC
d7278ab fix typo in param name 12 May 2016, 09:35:52 UTC
78642b6 add config generation for parallel sgd 12 May 2016, 09:35:51 UTC
744db6c add config generation for parallel sgd 12 May 2016, 09:35:50 UTC
8582774 add sgd parallel params 12 May 2016, 09:35:48 UTC
60daf7f guard against null values in test 12 May 2016, 09:35:47 UTC
037f150 add more unit tests for eval 12 May 2016, 09:35:46 UTC
3b80f20 consolidate global params 12 May 2016, 09:35:45 UTC
39e2d1b ExperimentalHTKMLFReader: add checkData option for filtering seq. of uneven length 12 May 2016, 08:43:46 UTC
5600f28 TruncatedBpttPacker.cpp: limit m_numParallelSequences for each worker 12 May 2016, 08:43:46 UTC
1d05742 Fixed GetDeviceId -> GetComputeDeviceId 11 May 2016, 23:36:58 UTC
b88cdc3 Fixes related to comments from the CR. 11 May 2016, 23:36:58 UTC
7816b5f Fixed transferFromDeviceToDevice 11 May 2016, 23:36:58 UTC
04041e9 Fixed a memory leak in GPUMatrix 11 May 2016, 23:36:58 UTC
3e70e43 Merge branch 'DanielMerget-bugfix_grayscale' into alexeyk/ir_fix 11 May 2016, 18:20:40 UTC
f603104 Refactored ImageReader config and fixed multiview10 as default bug. 11 May 2016, 18:12:51 UTC
f334cb9 bugfix: grayscale image now assumed by default if number of channels is 1 11 May 2016, 13:01:23 UTC
83cad6c Integrate mahilleb/RemoveUseValidationParameter into master 11 May 2016, 10:48:47 UTC
1a7ecae Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/seq2seqsample 10 May 2016, 17:19:48 UTC
5f0253b CR feedback for non-atomic reduction 10 May 2016, 16:51:04 UTC
33feb2c .cntk: remove obsolete useValidation= parameter 10 May 2016, 15:56:00 UTC
c15d9dd Tests/EndToEndTests/ModelExport/Model?/: limit to GPU SKU 10 May 2016, 14:11:38 UTC
56531ec Tests/EndToEndTests/CNTKTextFormatReader/Speech/Simple/: disabled Windows CPU Debug for nightly 10 May 2016, 14:11:21 UTC
ca03932 Integrate wilrich/logregacc into master 10 May 2016, 12:44:02 UTC
23f504c Enable clean up again 10 May 2016, 12:20:15 UTC
9541f38 Integrate wilrich/logregacc into master 10 May 2016, 12:18:02 UTC
6130eba Minor changes (error msg) 10 May 2016, 11:38:10 UTC
8ece70d merged 10 May 2016, 11:28:39 UTC
3f70c10 Merge branch 'wilrich/logregacc' of https://github.com/Microsoft/CNTK into wilrich/logregacc 10 May 2016, 11:27:49 UTC
b12f6f8 Aligning tol also for normal logreg 10 May 2016, 11:26:30 UTC
3f4a000 MLFDataDeserializer: change m_keyToSequence to vector; remove m_frames 10 May 2016, 10:21:26 UTC
6a88bcb changed reduction to 'double' (since we are not 100% identical anyway, we can make this switch); bug fix: an XCOPY command line missed a /D option 09 May 2016, 20:41:07 UTC
8f34f54 non-atomic reduction in TensorView 09 May 2016, 18:28:07 UTC
2a3bfd0 Integrate mahilleb/ExperimentalHTKMLFReservations into master 09 May 2016, 16:41:07 UTC
1539fea Integrate eldak/implCorpusDescriptor into master 09 May 2016, 16:24:16 UTC
f385553 ExperimentalHTKMLFReader: fix .reserve() for GetSequenceForChunk() 09 May 2016, 14:48:19 UTC
6e11c52 Aligning tol also for normal logreg 09 May 2016, 14:32:08 UTC
77109de Merge branch 'wilrich/logregacc' of https://github.com/Microsoft/CNTK into wilrich/logregacc 09 May 2016, 14:23:20 UTC
cb3f7ab Aligned the NumPy example with the file based one; make NumPy examples reproducible 09 May 2016, 14:20:45 UTC
f3329fc Aligned the NumPy example with the file based one 09 May 2016, 13:49:19 UTC
744cfb7 Integrate alrezni/text_reader_more_tests into master 09 May 2016, 12:57:35 UTC
6523da3 Add more tests for CNTKTextFormatReader * Add a couple of tests to ensure that randomize+read+pack does not garble any date in an epoch. The whole dataset in each test constitutes a single epoch, which is broken down into several large minibatches with each minibatch constructed form a number of randomly selected sequences. Each tests reads in the data in batches, packs them, and writes out packed batches to an output file. Then, the output file is sorted and is compared it against the original (sorted) dataset. * Refactor a number of functions in ReaderTestHelper 09 May 2016, 12:10:33 UTC
8e3b930 Fixing merge issue 09 May 2016, 07:58:04 UTC
e381e2a Adding missing std 09 May 2016, 07:58:04 UTC
3b42851 Adding overwrite option 09 May 2016, 07:58:04 UTC
bcc8410 Adding sanity check tests 09 May 2016, 07:58:04 UTC
673abd7 First implementation of the corpus descriptor 09 May 2016, 07:58:04 UTC
0a3c98c towards non-atomic two-step tensor reduction 08 May 2016, 02:29:02 UTC
7705b08 Updated Linux baselines 07 May 2016, 17:21:57 UTC
f26f6a7 Added dropout E2E test and fixed some dropout related bugs 07 May 2016, 01:58:35 UTC
5e0f3d7 bug fix: TensorView::AsMatrix() must determine needsReshaping correctly if needsSlicing 07 May 2016, 00:20:51 UTC
7981dd8 reenabled atomic reduction, for comparison in Jenkins 06 May 2016, 22:48:57 UTC
1d687b4 narrowing down non-reproducible tensor reduction 06 May 2016, 22:43:00 UTC
1dad7eb Merge branch 'master' of https://github.com/Microsoft/CNTK into dongyu/rollbackfix 06 May 2016, 18:54:42 UTC
bb71b49 Integrate fseide/seq2seqsample into master 06 May 2016, 18:31:40 UTC
96a3779 Merge branch 'master' of https://github.com/Microsoft/CNTK into dongyu/rollbackfix 06 May 2016, 17:41:22 UTC
8f6295e Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/dropoutRandomization 06 May 2016, 17:14:23 UTC
dbaee0c Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/seq2seqsample 06 May 2016, 17:02:06 UTC
2e6eadc two minor bug fixes in seq-2-seq samples 06 May 2016, 16:56:32 UTC
bd9b096 concepts documentation for row and col major, and examples 06 May 2016, 15:16:29 UTC
b58951c Integrate gaizna/cpp_eval_client_build_fix into master 06 May 2016, 10:01:23 UTC
7326be0 This commit fixes the project dependency order inside the CPPEvalClient.vcxproj file 06 May 2016, 08:12:23 UTC
ed75033 (cleaned up a comment) 06 May 2016, 07:25:24 UTC
b41df14 (missed a closing bracket...) 06 May 2016, 07:21:31 UTC
2b60c41 updated ExperimentalHtkmlfReader/LSTM test w.r.t. the stabilizer test 06 May 2016, 06:57:40 UTC
5203d32 addressed CR feedback 06 May 2016, 06:42:29 UTC
2952215 Addressed some CR feedback 06 May 2016, 06:26:25 UTC
43c62da (a comment edit) 06 May 2016, 04:18:42 UTC
7c7b1c4 reenabled atomic Scatter(); edited comments in CNTK.core.bs 06 May 2016, 04:16:29 UTC
c936a95 further update baselines. 06 May 2016, 04:04:36 UTC
267cafd Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/seq2seqsample 06 May 2016, 03:39:40 UTC
d757016 changed Stabilizer() to Softplus, which is more stable than the Exp version; changed seq-2-seq sample to TransposeTimes() implementation of ce and errs criteria 06 May 2016, 03:37:22 UTC
6be3394 fixed the last fix 06 May 2016, 03:33:49 UTC
5bc1f24 updated LSTM sample criterion to use SumElements() which reenables the old, buggy momentum computation; reenabled atomicAdd() reduction 06 May 2016, 03:32:26 UTC
6a4ecd9 LSTM went back to custom ce/err with TransposeTimes but no atomicAdd() 06 May 2016, 03:15:03 UTC
9e12c18 trying old criteria for lstm test 06 May 2016, 01:44:30 UTC
efa1372 and fixed the fix' fix 05 May 2016, 23:50:55 UTC
395d394 fixed the last fix 05 May 2016, 23:45:38 UTC
e5159cf rolled back e2e LSTM test to use the old stabilizer, for baseline compat 05 May 2016, 23:30:43 UTC
d75fdee addressed CR feedback 05 May 2016, 23:22:52 UTC
5f2fd9f addressed review feedback; made gcc happier 05 May 2016, 23:19:25 UTC
c2e9135 bug fix: Gather/Scatter() should also test for nan on CPU; made gcc happy 05 May 2016, 22:57:50 UTC
db8e480 Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/dropoutRandomization 05 May 2016, 22:51:43 UTC
b78cd13 switched LSTM E2E test config to use TransposeTimes() for criteria 05 May 2016, 22:05:14 UTC
fab6864 bug fix: TransposeTimes() should work with 1D tensors (=column vectors) 05 May 2016, 21:53:09 UTC
163a49b News in ReadMe. May 5, 2016 05 May 2016, 19:57:20 UTC
f4204ce Change in dropout randomization implementation to ensure reproducability across checkpoints 05 May 2016, 19:17:10 UTC
fd28c55 reviewed uses of atomicAdd(),; temporarily disallowed Scatter() reductions 05 May 2016, 18:02:21 UTC
back to top