Skip to main content

Browse the archive

https://github.com/Microsoft/CNTK
10 March 2023, 01:01:58 UTC
sort by:
Revision Author Date Message Commit Date
53fb3a7 Frank Seide 08 September 2016, 22:28:22 UTC merged from master; added a few TODO comments 08 September 2016, 22:28:22 UTC
392b6d3 Project Philly 08 September 2016, 19:43:42 UTC Integrate jiajia/WeigtedLogLossFix into master 08 September 2016, 19:43:42 UTC
49bcd34 Cha Zhang 08 September 2016, 17:15:18 UTC Minor revision (mostly in comments) after code review. 08 September 2016, 17:17:08 UTC
1d53a13 Cha Zhang 08 September 2016, 02:42:40 UTC Revised brainscript and code for correct glorotNorm initialization. Previous code computed fan-out wrong. Revision involved modifying the ParameterTensor claim, which also impacts the convolution CPU implementation. 08 September 2016, 17:17:08 UTC
0133303 Jian Jiao 08 September 2016, 16:58:29 UTC fixed weighted log loss 08 September 2016, 16:58:29 UTC
d0e7db5 Frank Seide 08 September 2016, 15:16:51 UTC (keep gcc happy) 08 September 2016, 15:16:51 UTC
0101e91 Willi Richert 08 September 2016, 14:36:44 UTC Set seed in FeedForwardNet example 08 September 2016, 14:36:44 UTC
44a998b Willi Richert 08 September 2016, 11:30:37 UTC Add python path dynamically; improved error handling 08 September 2016, 11:30:37 UTC
d2de39c yuxiao.guo 08 September 2016, 09:39:34 UTC Merge remote-tracking branch 'origin/master' into v-yuxgu/resnet-rc 08 September 2016, 09:39:34 UTC
49661bb yuxiaoguo 07 September 2016, 07:27:16 UTC Fix a bug in bn of our code This reverts commit d47adfbfeab311f95bbe04183c20109c918174f8. 08 September 2016, 08:48:31 UTC
d4e35ce Frank Seide 08 September 2016, 02:40:37 UTC DelayedValueNodeBase::BackpropTo() now does backprop into initial state as a bulk operation 08 September 2016, 02:40:37 UTC
d82f366 Frank Seide 08 September 2016, 01:54:34 UTC bug fix: back prop into initial state got conditions wrong 08 September 2016, 01:54:34 UTC
f4c2d14 Frank Seide 08 September 2016, 01:22:13 UTC DelayedValueNodeBase::GetMaskTensor() now uses the proper functions to do its thing 08 September 2016, 01:22:13 UTC
484656d Frank Seide 07 September 2016, 23:50:24 UTC streamlined DelayedValueNode mask computation 07 September 2016, 23:50:24 UTC
17bc7c7 Frank Seide 07 September 2016, 23:23:03 UTC rough cut of preparing DelayedValueNode masking in BeginForwardProp() 07 September 2016, 23:23:03 UTC
665eab0 Frank Seide 07 September 2016, 18:38:55 UTC bug fix in gap exclusion in DelayedValueNode 07 September 2016, 18:38:55 UTC
6908135 Frank Seide 07 September 2016, 16:58:44 UTC towards precomputing DelayedValueNodeBase's mask CPU-side in BeginForwardProp(); bug fix: more traceLevel fixes for other network builders 07 September 2016, 16:58:44 UTC
dda783d Project Philly 07 September 2016, 15:11:33 UTC Integrate zhouwang/remove-execonfig into master 07 September 2016, 15:11:33 UTC
27e65d4 Frank Seide 07 September 2016, 15:09:15 UTC intermediate step towards more efficient mask matrix DelayedValueNodeBase; bug fix: traceLevel should be read out when loading models from file and when creating them with old builders 07 September 2016, 15:09:15 UTC
1c32068 Zhou Wang 07 September 2016, 07:09:39 UTC remove unused solution folder 07 September 2016, 07:09:39 UTC
2545337 Frank Seide 07 September 2016, 06:06:28 UTC DelayerValueNodeBase now uses the same time offsets for backprop, in prep for sharing layouts; backprop into initial state seems to work now; addressed CR feedback 07 September 2016, 06:06:28 UTC
0ac275c Frank Seide 07 September 2016, 04:17:19 UTC (minor change in PastValue()) 07 September 2016, 04:17:19 UTC
0ab222b yuxiaoguo 07 September 2016, 02:57:41 UTC merge with master 07 September 2016, 02:57:41 UTC
a24c296 yuxiaoguo 07 September 2016, 02:15:31 UTC revert 1BitSGD 07 September 2016, 02:15:31 UTC
b909fc5 Frank Seide 07 September 2016, 02:09:21 UTC implemented backprop into DelayedValueNode's initialState 07 September 2016, 02:09:21 UTC
284778b Frank Seide 07 September 2016, 01:32:52 UTC DelayedValueNodeBase::Forward() now takes initial state from an external node (backprop into it still missing) 07 September 2016, 01:32:52 UTC
9a8e683 Frank Seide 06 September 2016, 18:07:10 UTC better error message in CloneFunction() 06 September 2016, 18:07:10 UTC
493c339 jeanfad 06 September 2016, 15:55:15 UTC add cntk_py to the tree 06 September 2016, 15:55:15 UTC
7beb59b jeanfad 06 September 2016, 15:19:44 UTC address cr comment 06 September 2016, 15:19:44 UTC
2b4dea7 jeanfad 06 September 2016, 15:08:53 UTC Merge branch 'blis/PythonV2' of https://github.com/Microsoft/CNTK into blis/PythonV2 06 September 2016, 15:08:53 UTC
5f9778b jeanfad 06 September 2016, 15:08:33 UTC example with numpy data 06 September 2016, 15:08:33 UTC
86e5e87 Amit 06 September 2016, 13:33:31 UTC CNTK v2 library: Python bindings related changes for linux 06 September 2016, 13:33:31 UTC
ea012e8 jeanfad 06 September 2016, 11:54:17 UTC Merge branch 'master' into blis/PythonV2 06 September 2016, 11:54:17 UTC
9229cf0 jeanfad 06 September 2016, 11:49:43 UTC resolve conflicts and address cr comments 06 September 2016, 11:49:43 UTC
7d5ca98 jeanfad 06 September 2016, 08:44:11 UTC shape.__getitem__ now revrses the dimensions for col row major 06 September 2016, 08:44:11 UTC
c37240a yuxiao.guo 06 September 2016, 06:31:43 UTC delete two more lines 06 September 2016, 06:31:43 UTC
8b46569 yuxiao.guo 06 September 2016, 02:31:02 UTC add more description on original ResNet 06 September 2016, 02:31:02 UTC
8cff694 Amit Agarwal 06 September 2016, 01:08:42 UTC CNTK v2 library: Sequence to sequence python example 06 September 2016, 01:11:27 UTC
994b500 Frank Seide 05 September 2016, 23:46:16 UTC optimized DelayedValueNode by using non-syncing mask 05 September 2016, 23:46:16 UTC
fa12c5e jeanfad 05 September 2016, 20:58:44 UTC remove duplicated printing method 05 September 2016, 20:58:44 UTC
e0352f9 jeanfad 05 September 2016, 20:57:33 UTC fix printing of loss 05 September 2016, 20:57:33 UTC
e55237d jeanfad 05 September 2016, 20:56:15 UTC fix dimensions in cifar example 05 September 2016, 20:56:15 UTC
6d564a5 jeanfad 05 September 2016, 20:51:17 UTC gix gpu problem when copying ndarrayview 05 September 2016, 20:51:17 UTC
34e3ee0 jeanfad 05 September 2016, 20:41:37 UTC change gpudevice in resnet file 05 September 2016, 20:41:37 UTC
c88fb38 jeanfad 05 September 2016, 20:38:49 UTC fix merge conflict 05 September 2016, 20:38:49 UTC
b3f2a0b jeanfad 05 September 2016, 20:34:47 UTC renaming of sgdlearner, cpu and gpu device, plus usage of default device and make funciton return output instead for replace_place holders 05 September 2016, 20:34:47 UTC
3b763f7 Amit Agarwal 05 September 2016, 20:22:51 UTC CNTK v2 library: Fixed some python wrappers 05 September 2016, 20:22:51 UTC
a423aa8 Eldar Akchurin 05 September 2016, 17:07:14 UTC Addressing CR 05 September 2016, 17:09:26 UTC
09d68bf Eldar Akchurin 05 September 2016, 17:02:32 UTC Fixing a bug 05 September 2016, 17:09:26 UTC
8f34c96 Eldar Akchurin 05 September 2016, 12:17:50 UTC Fixing baseline 05 September 2016, 17:09:26 UTC
527e472 Eldar Akchurin 05 September 2016, 10:37:38 UTC Fixing unavailable chunk in randomization 05 September 2016, 17:09:26 UTC
d610b2c jeanfad 05 September 2016, 16:42:27 UTC add todos related to variable 05 September 2016, 16:42:27 UTC
1580b23 jeanfad 05 September 2016, 16:23:24 UTC user facing improvements before the alpha release 05 September 2016, 16:23:24 UTC
f464f05 Zhou Wang 05 September 2016, 15:04:12 UTC remove CSEvalClientTest.exe.config and other eval test executables 05 September 2016, 15:24:11 UTC
6882490 Project Philly 05 September 2016, 14:21:00 UTC Integrate zhouwang/evalsdk2 into master 05 September 2016, 14:21:00 UTC
34310a7 jeanfad 05 September 2016, 13:06:57 UTC Merge branch 'amitaga/cntkv2LibraryOperators' of https://github.com/Microsoft/CNTK into amitaga/cntkv2LibraryOperators 05 September 2016, 13:06:57 UTC
12a0220 Amit Agarwal 05 September 2016, 05:54:53 UTC CNTK v2 library: Cleaned up python examples and added some inline documentation 05 September 2016, 13:04:56 UTC
fda75d9 yuxiao.guo 05 September 2016, 10:34:50 UTC change the folder name to OriResNet and rename of cntk file 05 September 2016, 10:34:50 UTC
d1b7d8c yuxiao.guo 05 September 2016, 10:30:16 UTC add the original ResNet scripits 05 September 2016, 10:30:16 UTC
8b646e5 Frank Seide 05 September 2016, 08:51:28 UTC bug fix in DelayedValueNodeBase::ForwardProp() 05 September 2016, 08:51:28 UTC
d0f6c8c yuxiao.guo 05 September 2016, 08:10:15 UTC revert the brace into a new line 05 September 2016, 08:10:15 UTC
61cf0f7 Frank Seide 05 September 2016, 08:08:58 UTC DelayedValueNodeBase::BackpropTo() now also uses a masked operation 05 September 2016, 08:08:58 UTC
484059e Frank Seide 05 September 2016, 07:28:17 UTC DelayedValueNodeBase::ForwardProp() cleaned up 05 September 2016, 07:28:17 UTC
c4db354 yuxiao.guo 05 September 2016, 07:25:33 UTC add a space after if 05 September 2016, 07:25:33 UTC
64b407e yuxiao.guo 05 September 2016, 07:19:38 UTC using space instead of tab 05 September 2016, 07:19:38 UTC
f13bef4 Frank Seide 05 September 2016, 06:59:38 UTC DelayedValueNodeBase::ForwardProp() now uses a masked operation 05 September 2016, 06:59:38 UTC
b9575aa Frank Seide 05 September 2016, 06:19:32 UTC changed DelayedValueNodeBase::ForwardProp() to use TensorView 05 September 2016, 06:19:32 UTC
78ac2a7 yuxiao.guo 05 September 2016, 06:15:59 UTC move the new resnet into a new handson file 05 September 2016, 06:15:59 UTC
441a970 Frank Seide 05 September 2016, 06:00:02 UTC refactored DelayedValueNodeBase::ForwardProp() 05 September 2016, 06:00:02 UTC
64e8eff Amit Agarwal 05 September 2016, 05:54:53 UTC CNTK v2 library: Cleaned up python examples and added some inline documentation 05 September 2016, 05:54:53 UTC
456863b Frank Seide 05 September 2016, 01:43:07 UTC DelayedValueNodeBase::BackpropTo() now uses TensorView, in prep for using a mask 05 September 2016, 01:43:07 UTC
e9037ce Frank Seide 05 September 2016, 01:36:17 UTC added RecurrentNodes.cpp to Linux Makefile 05 September 2016, 01:36:17 UTC
5838b2a Frank Seide 05 September 2016, 01:27:24 UTC optimized DelayedValueNodeBase::ForwardProp() 05 September 2016, 01:27:24 UTC
aa664df Frank Seide 04 September 2016, 23:58:00 UTC fixed indentation of RecurrentNodes.cpp 04 September 2016, 23:58:00 UTC
5000df7 Frank Seide 04 September 2016, 23:54:45 UTC sorted method definitions in RecurrentNodes.cpp 04 September 2016, 23:54:45 UTC
d006a5f Frank Seide 04 September 2016, 23:50:55 UTC completed the split off RecurrentNodes into a CPP 04 September 2016, 23:50:55 UTC
bc58c5c Frank Seide 04 September 2016, 23:09:07 UTC (first step of moving RecurrentNodes to a CPP file) 04 September 2016, 23:09:07 UTC
35c4422 Frank Seide 04 September 2016, 22:56:55 UTC bug fixes: traceLevel parameters must be treated as optional; fixed a left-over %zd 04 September 2016, 22:56:55 UTC
37a236b Frank Seide 04 September 2016, 06:31:46 UTC (made gcc happy) 04 September 2016, 06:31:46 UTC
84cd533 Frank Seide 04 September 2016, 06:05:35 UTC fixed the incorrect minibatch range for the firstMBsToShowResults 04 September 2016, 06:05:35 UTC
a5bbc86 Frank Seide 04 September 2016, 05:50:08 UTC further minor refinements (spaces, newlines, spelling) in SGD logging, esp. minibatch tracing 04 September 2016, 05:50:08 UTC
df597ce Frank Seide 04 September 2016, 05:16:13 UTC further guarding some log spew by traceLevel>0 04 September 2016, 05:16:13 UTC
c72fa31 Frank Seide 04 September 2016, 03:58:14 UTC log spew in training and model creation/loading now guarded by traceLevel>0; log pathname is now just the stderr parameter; log path "rankN" is now ".rankN" and not added for rank 0 04 September 2016, 03:58:14 UTC
fb41fbd Amit Agarwal 04 September 2016, 00:38:23 UTC CNTK v2 library: Address linux build issue 04 September 2016, 00:38:23 UTC
3f9f248 Amit Agarwal 04 September 2016, 00:31:11 UTC Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/cntkv2LibraryOperators 04 September 2016, 00:31:11 UTC
abf4680 Amit Agarwal 04 September 2016, 00:31:02 UTC CNTK v2 library: Trainer checkpointing and some MinibatchSource creation helper APIs 04 September 2016, 00:31:02 UTC
301880c jeanfad 03 September 2016, 17:19:40 UTC readme file update 03 September 2016, 17:19:40 UTC
b20f921 Jean Baptiste Faddoul 03 September 2016, 17:04:42 UTC typo fix 03 September 2016, 17:04:42 UTC
f1af5ec Frank Seide 03 September 2016, 06:59:15 UTC (fixed log messages in MB scaling) 03 September 2016, 06:59:15 UTC
e7a45fb Frank Seide 03 September 2016, 06:28:28 UTC Merge branch 'master' of https://github.com/Microsoft/cntk into fseide/samplebs 03 September 2016, 06:28:28 UTC
179ba7f Frank Seide 03 September 2016, 06:05:29 UTC changed fsAdaGrad magic constant from 1/400 to 1, as it seems to work better. Now it quite closely resembles the popular Adam optimizer 03 September 2016, 06:05:29 UTC
f1a0f0d Frank Seide 03 September 2016, 05:52:13 UTC bug fix: FSAdagrad denominator really should not be kept in a static variable and double-counted; bug fix: LSTMBlock{} should not used lambdas for its parameters 03 September 2016, 05:52:13 UTC
6988e36 Frank Seide 03 September 2016, 05:44:20 UTC fsAdaGrad smoothedCounts array created and passed around with smoothedGradients (but not used yet) 03 September 2016, 05:44:20 UTC
6062b7b Frank Seide 03 September 2016, 05:22:07 UTC inlined old UpdateWeights(), renamed UpdateWeightsS() to UpdateWeights() 03 September 2016, 05:22:07 UTC
518baf6 Frank Seide 03 September 2016, 05:10:06 UTC fsAdaGrad magic constants exposed as parameters; InitRandom() now shows the first random number, so one can discover changes from the log 03 September 2016, 05:10:06 UTC
0f4ca24 Frank Seide 03 September 2016, 04:50:06 UTC added traceLevel to ComputationNetwork (through ComputationEnvironment); added exhaustive forward/backprop logging for extreme debugging, under traceLevel=1000000; bug fix in WriteMinibatchWithFormatting(): must show all elements also for matrices without MBLayout; refactored FSAdaGrad, towards fixing the "static" bug (still repros the bug); removed an unnecessary warning; fix in RecurrentLSTMLayerStack{} 03 September 2016, 04:50:06 UTC
fcd66a9 Eldar Akchurin 02 September 2016, 19:57:04 UTC Updating the baseline 02 September 2016, 19:58:15 UTC
921f141 Eldar Akchurin 02 September 2016, 15:06:52 UTC Updating baseline 02 September 2016, 19:58:15 UTC
ef2cac4 Eldar Akchurin 02 September 2016, 14:17:46 UTC Removing unnecessary parenthesis 02 September 2016, 19:58:15 UTC
743a595 Eldar Akchurin 02 September 2016, 14:15:53 UTC Adding more unit test and fixing randomization 02 September 2016, 19:58:15 UTC
back to top