89e0958 | Mark Hillebrand | 18 January 2016, 08:33:34 UTC | License change | 18 January 2016, 08:33:34 UTC |
602c37a | Frank Seide | 08 September 2015, 22:35:34 UTC | Merge branch 'master' into fseide/netlib Conflicts: MachineLearning/CNTK/CNTK.cpp | 08 September 2015, 22:35:34 UTC |
806e184 | Amit Agarwal | 08 September 2015, 20:24:04 UTC | Updated Windows baseline files for Speech/LSTM test and also undid the recent increase in tolerance % for the test | 08 September 2015, 20:24:04 UTC |
941c7d7 | Amit | 08 September 2015, 20:06:21 UTC | Updated Linux baselines for the LSTM test to accomodate the change in parameter initialization | 08 September 2015, 20:06:21 UTC |
c085b4b | Amit | 08 September 2015, 19:23:18 UTC | Fixed a crash on Linux | 08 September 2015, 19:23:18 UTC |
c62c813 | Frank Seide | 06 September 2015, 19:47:30 UTC | added ParallelTraining test cases to the VS Solution; relaxed baseline for single precision since now all scalars are double precision, which changes the result (double-precision run should not change) | 06 September 2015, 19:47:30 UTC |
5ef77ad | Frank Seide | 06 September 2015, 19:23:23 UTC | Merge branch 'fseide/netlib' of https://git.codeplex.com/cntk into fseide/netlib | 06 September 2015, 19:23:23 UTC |
0b06745 | Frank Seide | 06 September 2015, 19:17:37 UTC | added Linux test cmd to README; added dllexport to DeviceFromConfig() | 06 September 2015, 19:17:37 UTC |
50ff0eb | Frank Seide | 06 September 2015, 19:10:15 UTC | fixed CNTKMathTest (includes have moved) | 06 September 2015, 19:10:15 UTC |
89880de | Frank Seide | 06 September 2015, 18:55:51 UTC | somehow screwed up lib directory path for CNTKEval.dll--fixed | 06 September 2015, 18:55:51 UTC |
b6378ec | Frank Seide | 06 September 2015, 16:29:07 UTC | missed stdafx etc in repo | 06 September 2015, 16:29:07 UTC |
54a6b1d | Frank Seide | 06 September 2015, 16:20:28 UTC | made gcc happy again (mostly missing headers or wrong declaration orders); Makefile adapted to new paths, but not yet building Network and SGD as separate libs | 06 September 2015, 16:20:28 UTC |
0a67d0c | Frank Seide | 06 September 2015, 15:21:24 UTC | Merge branch 'master' of https://git.codeplex.com/cntk into fseide/netlib | 06 September 2015, 15:21:24 UTC |
d4dce0c | Frank Seide | 06 September 2015, 15:20:30 UTC | (deleted some left-over) | 06 September 2015, 15:20:30 UTC |
27d12d2 | Frank Seide | 06 September 2015, 15:17:07 UTC | Matrix.h now no longer pulls in the CPU/GPUMatrix headers (this required cleaning up a few incorrect header dependencies as well); cleaned up <class ElemType> vs. <typename ElemType> (using class since that was used more often, causing less diffs) | 06 September 2015, 15:17:07 UTC |
580d2c4 | Frank Seide | 06 September 2015, 04:19:34 UTC | fixed a linker warning caused by BestGpu.cpp in CPUONLY mode saying it does not export any symbols | 06 September 2015, 04:19:34 UTC |
08d5298 | Frank Seide | 06 September 2015, 04:14:47 UTC | added comments on further disentangling this for BrainScript and Matrix; updated the post-build event of CNTKEval to not fail in CPU-only builds | 06 September 2015, 04:14:47 UTC |
02b5101 | Frank Seide | 05 September 2015, 15:28:27 UTC | InitLearnableParametersFromFile() moved out of ComputationNetwork into LearnableParameter itself, there called InitFromFile(); renamed LoadArrayFromFile() to LoadMatrixFromFile() since that's what it does and means, even if it passes it on temporarily as a flattened array | 05 September 2015, 15:28:27 UTC |
d8e5660 | Frank Seide | 05 September 2015, 14:30:19 UTC | moved LoadArrayFromTextFile() from ComputationNetwork.h to File.h (it fits only a little better there) | 05 September 2015, 14:30:19 UTC |
93d7883 | Frank Seide | 05 September 2015, 12:11:58 UTC | fixed CNTKEval after the last changes | 05 September 2015, 12:11:58 UTC |
07ac709 | Frank Seide | 05 September 2015, 11:57:27 UTC | (comments) | 05 September 2015, 11:57:27 UTC |
e61635a | Frank Seide | 05 September 2015, 11:49:40 UTC | moved DecimateMinibatch functions to CPP | 05 September 2015, 11:49:40 UTC |
1421a87 | Frank Seide | 05 September 2015, 10:05:36 UTC | renamed a VS filter and added a few comments | 05 September 2015, 10:05:36 UTC |
73e025b | Frank Seide | 05 September 2015, 09:28:24 UTC | moved IComputationNetBuilder.h over to CNTKSGDLib, as it is the interface through which models are either created or loaded from check-point | 05 September 2015, 09:28:24 UTC |
1096393 | Frank Seide | 05 September 2015, 09:18:44 UTC | renamed CNTKEval and CNTKMath by appending -Dll to their name | 05 September 2015, 09:18:44 UTC |
da176ea | Frank Seide | 05 September 2015, 09:16:55 UTC | CNTK now delay-loads msmpi.dll in case it is not installed; moved Profiled.cpp/h; disabled MultiNetworksSGD--seems not missing, not used | 05 September 2015, 09:16:55 UTC |
7c5e4d3 | Frank Seide | 05 September 2015, 09:00:51 UTC | moved stuff out from CNTK to CNTKSGDLib; created an SGD.cpp that instantiates the exported classes of CNTKSGDLib; does not build since git mixed up files during move, and it won't let me git-add and git-mv in one go | 05 September 2015, 09:00:51 UTC |
419d6af | Frank Seide | 05 September 2015, 08:11:45 UTC | fixed project files after moving node/network sources over; forgot to git-add the new file of previous check-in | 05 September 2015, 08:11:45 UTC |
a6d2d0a | Frank Seide | 05 September 2015, 08:03:40 UTC | moved sources to CNTKComputationNetworkLib (will fix VS projects next) | 05 September 2015, 08:03:40 UTC |
fc938ec | Frank Seide | 05 September 2015, 07:56:40 UTC | split ExperimentalNetworkBuilder.cpp up by moving out the actual node/network-creation code (which does not need to know ExperimentalNetworkBuilder.h) to new file NetworkBuilderFromConfig.cpp | 05 September 2015, 07:56:40 UTC |
8ac69d4 | Frank Seide | 05 September 2015, 07:44:16 UTC | updated CNTKEval w.r.t. network lib | 05 September 2015, 07:44:16 UTC |
0d658d0 | Frank Seide | 05 September 2015, 07:40:20 UTC | removed all CPP files of ComputationNetwork from CNTK project | 05 September 2015, 07:40:20 UTC |
15bcde7 | Frank Seide | 05 September 2015, 07:35:44 UTC | changed XXX<float>::TypeName() to OperationNameOf(XXX) | 05 September 2015, 07:35:44 UTC |
f061291 | Frank Seide | 05 September 2015, 07:21:53 UTC | changed several scalar values from ElemType to double, including all objective values, error metrics, learning rates, dropout rate, adaptation weights etc., to eliminate more <ElemType> depnedency; DistGradHeader no longer depending on <ElemType>; all accesses of ComputationNode::TypeName are now done to <float> variant instead of <ElemType>, for consistency where we don't have an <ElemType> | 05 September 2015, 07:21:53 UTC |
95f2e11 | Frank Seide | 05 September 2015, 06:05:49 UTC | and deleted ComputationNetworkHelper.h--one more down! | 05 September 2015, 06:05:49 UTC |
f79f373 | Frank Seide | 05 September 2015, 06:04:00 UTC | eliminated ComputationNetworkHelper--it was a base class that contained a few evaluation helpers that support forward prop, they may equally well belong into ComputationNetwork directly; also, made two of three independent of ElemType | 05 September 2015, 06:04:00 UTC |
f3f0bc1 | Frank Seide | 05 September 2015, 05:39:46 UTC | now linking with CNTKComputationNetworkLib, removed one CPP from CNTK proper for testing | 05 September 2015, 05:39:46 UTC |
76a427e | Frank Seide | 05 September 2015, 03:12:24 UTC | selected which files go into the network lib | 05 September 2015, 03:12:24 UTC |
18f4305 | Jasha Droppo | 05 September 2015, 02:53:41 UTC | Revert "Win32 CNTK.exe added control-c handler to wrap checkpoint file writing. If checkpoint file gets corrupted, then entire training run may be ruined." This reverts commit 9b212d7783f0af010c4652ae87dba5b4b9730bea. | 05 September 2015, 02:53:41 UTC |
14ae5f3 | Frank Seide | 05 September 2015, 02:42:56 UTC | added a new projec to hold computation-network related stuff as a library, not yet an actual library | 05 September 2015, 02:42:56 UTC |
9b212d7 | Jasha Droppo | 05 September 2015, 02:31:00 UTC | Win32 CNTK.exe added control-c handler to wrap checkpoint file writing. If checkpoint file gets corrupted, then entire training run may be ruined. | 05 September 2015, 02:35:11 UTC |
afcdd44 | Vladimir Ivanov | 05 September 2015, 02:12:35 UTC | Restored proper permissions for run-test | 05 September 2015, 02:12:47 UTC |
9650981 | Vladimir Ivanov | 05 September 2015, 02:05:14 UTC | Fix TestDriver script permissions... | 05 September 2015, 02:05:14 UTC |
f0602a1 | Marko Radmilac | 04 September 2015, 19:22:12 UTC | Fix configure script permissions... again | 04 September 2015, 19:22:12 UTC |
12ed4a2 | Frank Seide | 04 September 2015, 04:18:13 UTC | made ComputationNetwork independent of <ElemType>, aiming to make it easier to expose this as an interface in the future. In particular, all methods that still depended on <ElemType> got their own template parameter; some spelling errors renamed (RequirePreCompute() -> RequirePreCompute(), same for BatchMode, and similarly for GetNodesRequiringPreComputation()) | 04 September 2015, 04:18:13 UTC |
2e628cb | Frank Seide | 04 September 2015, 02:35:17 UTC | merged fseide/bs feature branch: Added a completely new configuration language, which currently can be used in place of NDL, but eventually will power all configurations. It supports infix expressions, recursive macros, arrays, and a few useful functions such as string replace. It is called "BrainScript" (file extension .bs), where the name is meant to be reflective of our grand ambition (whereas the file extension is reflective of where we stand today w.r.t. that grand ambition...). As of now, BrainScript can be accessed for configuring networks through the new ExperimentalNetworkBuilder option. A few ComputationNodes are still missing, and MEL may not work as node naming is not sorted out yet. The core classes were refactored aiming at removing the pervasive template parameter <ElemType> (selecting float vs. double), aiming at making it feasible to wrap parts of CNTK as libraries. ComputationNode has been disentanlgled, while consumers such as ComputationNetwork and SGD--which really should be agnostic to float/double--have been changed to use the agnostic interface (ComputationNodeBase) where possible, but the full separation will require many more steps. Theoretically, once this is completed, it would be possible to mix float and double nodes in a single graph (through the use of still to be written typecast nodes). The two variants of each Evaluate and ComputePartial have been unified across full-minibatch and per-frame operation through passing the range as a new FrameRange object that encodes both whether it is the full minibatch vs. single frame, as well as the number of slices in a minibatch. Currently, the latter is passed through a member m_samplesInRecurrentStep, which now can be removed (currently it is kept for a runtime check to verify that this was done right--to be removed). The LSTM test case was modified to initialize its parameters with CPU code that, unlike the GPU code, honors random seeds, making it resilient to evaluation-order changes (that BrainScript implies, for example). The test case now has a BrainScript implementation (it is not default though; default remains the NDL version). Further minor code refactoring. | 04 September 2015, 02:35:17 UTC |
8e1c700 | Frank Seide | 04 September 2015, 02:23:21 UTC | few more gcc issues (that compiler is soo slow!!) | 04 September 2015, 02:23:21 UTC |
a4c32ac | Frank Seide | 04 September 2015, 02:15:00 UTC | CNTKEval now builds (missed the new .cpp files) | 04 September 2015, 02:15:00 UTC |
781c78e | Frank Seide | 04 September 2015, 02:06:17 UTC | few more gcc issues | 04 September 2015, 02:06:17 UTC |
af40073 | Frank Seide | 04 September 2015, 02:02:15 UTC | fixed more stuff broken under gcc | 04 September 2015, 02:02:15 UTC |
cb5621c | Frank Seide | 04 September 2015, 01:57:58 UTC | fixed a few gcc build problems, requiring more stuff to move from .h to .cpp | 04 September 2015, 01:57:58 UTC |
ef0f197 | Frank Seide | 04 September 2015, 01:47:41 UTC | somehow git forgot to add SynchronousExecutionEngine.cpp | 04 September 2015, 01:47:41 UTC |
3066a16 | Frank Seide | 04 September 2015, 01:41:59 UTC | moved most <ElemType>-dependent functions out of ComputationNetwork into a ComputationNetworkBuilder class (added two files); fixed the slew of dependencies, mostly SimpleNetworkBuilder.cpp; ComputationNetwork no longer has to include all headers for all vairants of ComputationNodes; a few ComputationNode derivates missed #includes | 04 September 2015, 01:41:59 UTC |
31c50cb | Frank Seide | 03 September 2015, 22:21:26 UTC | another fix related to the previous one | 03 September 2015, 22:21:26 UTC |
3e86195 | Frank Seide | 03 September 2015, 22:18:17 UTC | fixed wrong type of Matrix<float> in SetSentenceSegBatch(); fixed a missing percentage in testcases.yml | 03 September 2015, 22:18:17 UTC |
1219c24 | U-FAREAST\fseide | 03 September 2015, 22:05:46 UTC | Merge branch 'master' into fseide/bs | 03 September 2015, 22:05:46 UTC |
83ddaf0 | Frank Seide | 03 September 2015, 21:47:34 UTC | fixed a silly refactoring bug | 03 September 2015, 21:47:34 UTC |
a4c480f | Frank Seide | 03 September 2015, 20:50:03 UTC | Merge branch 'master' into fseide/bs Conflicts: Math/Math/NoGPU.cpp | 03 September 2015, 20:50:03 UTC |
d4b735a | Frank Seide | 03 September 2015, 20:44:27 UTC | fixed gcc template strictness issue in ExperimentalNetworkBuilder.cpp--thanks to Marko Radmiac for showing my feeble brain how to do that; bug fix: ComputationNode::OurElemType was incorrectly set to float instead of ElemType... ahem! increased eval err tolerance of LSTM test to 3%, should fix this some time by updating the baseline | 03 September 2015, 20:44:27 UTC |
4bbcca4 | Marko Radmilac | 03 September 2015, 19:17:17 UTC | NVCC warnings as errors | 03 September 2015, 20:30:30 UTC |
db08c5c | Frank Seide | 03 September 2015, 20:14:46 UTC | fixed two remaining build problems with gcc of the last commit | 03 September 2015, 20:14:46 UTC |
16bb5f9 | Frank Seide | 03 September 2015, 18:56:43 UTC | split ComputationNode into a part depending on <ElemType> and part not depending on it (called ComputationNodeBase); and changed consumers of this, such as ComputationNetwork and SGD, now, wherever possible, to operate on the independent item. This is aimed at allowing abstract interfaces into CNTK without the complexity of the <ElemType> type parameterization (which is unnecessary on the interface level); few ComputationNode implementations were adapted to consistently use Inputs() instead of m_children, since m_children[] will now give the base class; changes all sentence-boundary flag matrices from Matrix<ElemType> to Matrix<float> (hoping that one day we can use Matrix<enum type>). BUGBUG: the old LSTM node currently is now broken because it directly multiplies data with this matrix--not fixed since that node is not in use currently; changed LSTM test case tolerance to 2% to be resilient to changes in initial randomization; MatrixPool no longer depends on <ElemType>, but instead implements both a float and a double pool | 03 September 2015, 18:56:43 UTC |
4e32c0e | Marko Radmilac | 02 September 2015, 21:14:03 UTC | Fix Windows NVCC warnings | 03 September 2015, 17:17:07 UTC |
fe3346b | Marko Radmilac | 02 September 2015, 19:25:13 UTC | Cleanup warnings in Linux | 03 September 2015, 16:24:03 UTC |
329f2d1 | Frank Seide | 02 September 2015, 23:41:57 UTC | replaced lots of ColumnSlice() with FrameSlice() (but not all yet, to be completed later) | 02 September 2015, 23:41:57 UTC |
ff9c672 | Frank Seide | 02 September 2015, 21:57:57 UTC | first attempt with FrameSlice(), succeeded | 02 September 2015, 21:57:57 UTC |
4781287 | Frank Seide | 02 September 2015, 21:35:41 UTC | Merge branch 'fseide/framerange' into fseide/bs Conflicts: MachineLearning/CNTK/ComputationNode.h Tests/Speech/LSTM/cntk.config | 02 September 2015, 21:35:41 UTC |
e5cc613 | Frank Seide | 02 September 2015, 21:21:55 UTC | bug fix in FrameRange::IsAllFrames(), got the logic inverted; README.txt updated to match updated LSTM test case | 02 September 2015, 21:21:55 UTC |
1849d5d | U-FAREAST\fseide | 02 September 2015, 21:07:05 UTC | Merge branch 'master' into fseide/framerange | 02 September 2015, 21:07:05 UTC |
9e627a4 | Frank Seide | 02 September 2015, 21:06:10 UTC | FrameRange now not only takes a time index but also the number of slices in the minibatch, aiming to eliminate m_samplesInRecurrentStep; RecurrentNode and Past/FutureValueNodes now pass frameRange deeper down, eliminating explicit passing of m_samplesInRecurrentStep (which is already included in FrameRange); ComputeInputPartial() and EvaluateThisNode() implemented the same thing differently, changed to use identical pattern | 02 September 2015, 21:06:10 UTC |
035e164 | Amit | 02 September 2015, 20:09:18 UTC | Merge branch 'master' of https://git.codeplex.com/cntk into dongyu/LSTMTestCase | 02 September 2015, 20:09:18 UTC |
ceffaea | Amit | 02 September 2015, 20:09:07 UTC | Added platform specific baselines for the LSTM speech test | 02 September 2015, 20:09:07 UTC |
995fc87 | Frank Seide | 02 September 2015, 18:34:42 UTC | moved FrameRange to Matrix.h; added new function FrameSlice() which is like ColumnSlice() but with knowledge of frame semantics through the FrameRange object | 02 September 2015, 18:34:42 UTC |
86faad3 | Amit Agarwal | 02 September 2015, 18:19:44 UTC | Fixed a bug in utterance mode reading in HTKMLFReader that existed in the Linux fork and was propagted to the common reader with the merge of the readers | 02 September 2015, 18:19:44 UTC |
0bb5419 | Chris Basoglu | 02 September 2015, 17:35:32 UTC | Merge branch 'dongyu/LSTMTestCase' of https://git.codeplex.com/cntk into dongyu/LSTMTestCase Conflicts: Math/Math/Matrix.cpp | 02 September 2015, 17:35:32 UTC |
15b941e | Chris Basoglu | 02 September 2015, 15:12:27 UTC | Make LSTM/run-test have executable permissions for Linux | 02 September 2015, 17:32:11 UTC |
5ae107b | Dong Yu | 02 September 2015, 09:44:35 UTC | remove last four lines in LSTM test case's run-test so that it only runs once. | 02 September 2015, 17:32:11 UTC |
e63e215 | Dong Yu | 01 September 2015, 00:03:11 UTC | add NDLDir to run-test for the LSTM test case. | 02 September 2015, 17:32:11 UTC |
55f4f1d | Dong Yu | 31 August 2015, 23:18:48 UTC | finish LSTM test case | 02 September 2015, 17:32:05 UTC |
8705dd0 | Frank Seide | 02 September 2015, 17:12:15 UTC | Merge remote-tracking branch 'origin' into fseide/bs Conflicts: DataReader/HTKMLFReader/ssematrix.h Math/Math/CUDAPageLockedMemAllocator.cpp Math/Math/Matrix.cpp Math/Math/NoGPU.cpp Math/Math/ValueQuantizer.h | 02 September 2015, 17:12:15 UTC |
67464d2 | Frank Seide | 02 September 2015, 15:43:31 UTC | some fixes to make gcc happy (one still missing); somehow git lost 'Makefile'--readding, hoping it won't confuse the merge process | 02 September 2015, 15:43:31 UTC |
448d7a1 | Chris Basoglu | 02 September 2015, 15:12:27 UTC | Make LSTM/run-test have executable permissions for Linux | 02 September 2015, 15:12:27 UTC |
d8d5525 | Dong Yu | 02 September 2015, 09:44:35 UTC | remove last four lines in LSTM test case's run-test so that it only runs once. | 02 September 2015, 09:44:35 UTC |
86815cc | Frank Seide | 02 September 2015, 07:43:15 UTC | made the LSTM sample look more BS-like | 02 September 2015, 07:43:15 UTC |
c980d9f | Frank Seide | 02 September 2015, 06:59:26 UTC | bug fix in parser: could not skip to end if already at end (0 chars), e.g. // at line end; bug fix: .* is not DiagTimes but ElementTimes; BS LSTM sample further refined, still working well | 02 September 2015, 06:59:26 UTC |
cf0d15a | Amit | 02 September 2015, 05:26:37 UTC | Enable unigram support on Linux | 02 September 2015, 06:11:13 UTC |
67f5121 | Frank Seide | 02 September 2015, 06:02:04 UTC | using self-New macros in BS LSTM sample | 02 September 2015, 06:02:04 UTC |
b1ccf91 | Amit Agarwal | 02 September 2015, 05:07:15 UTC | Replace uses of the VC++ non-standard std::hash_map with C++ standard std::unordered_map type in the HTKMLFReader | 02 September 2015, 05:07:15 UTC |
6f0526e | Amit Agarwal | 02 September 2015, 01:23:17 UTC | 1) Switch to using std::mutex for mutual exclusion instead of platform specific facilities 2) A minor bug fix in the linux reader | 02 September 2015, 01:23:17 UTC |
2d1030c | Amit Agarwal | 01 September 2015, 23:01:19 UTC | Merge branch 'master' of https://git01.codeplex.com/cntk into amitaga/mergeHTKMLFReaders | 01 September 2015, 23:01:19 UTC |
48d5968 | Amit Agarwal | 01 September 2015, 23:01:05 UTC | Fixed a bug introduced during merging of the Linux and Windows readers | 01 September 2015, 23:01:05 UTC |
9773408 | Vladimir Ivanov | 01 September 2015, 22:13:28 UTC | Reporting duration of test runs | 01 September 2015, 22:13:28 UTC |
6c61b87 | Frank Seide | 01 September 2015, 22:08:47 UTC | started to be more fancy with BS | 01 September 2015, 22:08:47 UTC |
7ca0057 | Frank Seide | 01 September 2015, 21:38:27 UTC | added tracing of stack pointers to track stack usage--1 KB for Evaluate(), why?? temporarily increased stack allocation; LSTM test script now runs, and correctly !!!, with BS | 01 September 2015, 21:38:27 UTC |
cc1f6b3 | Frank Seide | 01 September 2015, 21:04:03 UTC | defined __declspec_noreturn for __declspec(noreturn) (empty on Linux); Basics.h now includes Plattform.h for __declspec_noreturn | 01 September 2015, 21:04:03 UTC |
d4c3e4a | Frank Seide | 01 September 2015, 20:44:20 UTC | added BS to Linux Makefile--with lots of fallout that needs to be fixed | 01 September 2015, 20:44:20 UTC |
8e85f07 | Amit Agarwal | 01 September 2015, 20:27:19 UTC | Merge branch 'master' of https://git01.codeplex.com/cntk into amitaga/mergeHTKMLFReaders | 01 September 2015, 20:27:19 UTC |
8087de0 | Frank Seide | 01 September 2015, 20:19:41 UTC | changed a lot of 'let' inside Evaluate() to 'let &' to save some stack space; changed lots of function arguments from shared_ptrs passed by value to passed as const &, to save ref counting overhead | 01 September 2015, 20:19:41 UTC |
1df4e8e | Amit Agarwal | 31 August 2015, 18:55:28 UTC | Merge the linux and windows forks of the HTKMLFReader fork into one. | 01 September 2015, 20:09:57 UTC |
e74e5e3 | Frank Seide | 01 September 2015, 19:17:06 UTC | #if-0'ed out the old fake objects that developed this along, since they caused an 'unreachable code' warning=error; decorated ConfigParser.cpp's Fail() functions as static noreturn as they should be | 01 September 2015, 19:17:06 UTC |