98a723c85b
reduced clutter of validation output by validating the entire network, rather than each output node individually
2016-02-04 14:51:53 -08:00
a930d1cfb4
Updated EvalActions to fix support for OutputNodeNames in the config.
2016-02-04 13:11:13 -08:00
14881a210e
Added guards to all model saving calls to prevent mutiple parallel workers from trying to write the model file concurrently
2016-02-04 10:55:28 -08:00
1e65a415cb
fixed bug in catching when MPI runtime is missing
2016-02-04 08:51:08 -08:00
8b2c9ffb53
Integrate mahilleb/ImageReaderOptDepFix into master
2016-02-04 04:22:24 -08:00
29dfe2d2aa
Integrate eldak/addressingReaderCR into master
2016-02-04 03:28:46 -08:00
7914c7c3d3
Integrate pkranen/fixMNIST into master
2016-02-04 03:13:07 -08:00
5fa84b0fb2
ImageReader.vcxproj: don't build if OpenCV not present
2016-02-04 11:00:26 +01:00
f5a349a11a
Addressing ImageReader CR from Frank
2016-02-04 09:37:14 +01:00
7d1fc361a9
fixed .config to .cntk in CSEvalCient
2016-02-04 09:11:06 +01:00
27fd4e045b
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations
2016-02-03 23:05:16 -08:00
eb231c37aa
(added a comment)
2016-02-03 16:52:47 -08:00
0900bdb928
gave up back-compat with some mis-spelled names of training criterion nodes
2016-02-03 15:27:00 -08:00
78587f586e
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations
2016-02-03 12:19:54 -08:00
070006d647
addressed an issue discovered in static code analysis
2016-02-03 11:49:13 -08:00
2d1198071c
addressed bugs discovered by static code analysis, Issue #81 , and manually merged two related pull requests
2016-02-03 11:32:13 -08:00
c3948330c5
Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixCrossProcessMutex
2016-02-03 09:32:47 -08:00
9b1f342aee
Integrate mahilleb/WindowsVsDependenciesDoc into master
2016-02-03 09:30:22 -08:00
87acb14f55
final simplification of case-insensitive string comparisons
2016-02-03 09:24:09 -08:00
901f99f5de
further simplified case-insensitive string compares
2016-02-03 09:17:58 -08:00
16175a17c5
cleanedup all case-insensitive comparisons
2016-02-03 09:01:55 -08:00
f4549625d1
fixed process-name comparison in BestGpu::QueryNvmlData()
2016-02-03 08:38:24 -08:00
7ab3d9e909
.vcxproj: Let diagnostics about dependencies point to CNTK Wiki
2016-02-03 17:22:13 +01:00
ee97506e1f
Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixCrossProcessMutex
2016-02-03 08:09:37 -08:00
fe75bb1650
Extend build information included / reported by CNTK
...
* Build target: GPU or CPU-only
* Build type (Debug or Release), was missing for Windows
* Whether or not 1bit-SGD was included
* If there were local modifications in the Git working copy
* Generate BUILD{ER,MACHINE,PATH} on Linux
2016-02-03 15:48:39 +01:00
d630aec812
change needsGradient to learningRateMultiplier. NDL and MEL has been changed, however BrainScript has not yet.
2016-02-02 17:59:54 -08:00
ba49d13661
removed more PairNode leftovers; bug fix in QueryNvmlData()
2016-02-02 17:24:54 -08:00
c21b2335e4
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations
2016-02-02 16:27:32 -08:00
36b174c242
Merge branch 'dongyu/BPTTWithContexts' of https://github.com/Microsoft/CNTK
2016-02-02 16:19:11 -08:00
8c38137967
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations
2016-02-02 15:58:36 -08:00
a3acfd6e53
Merge branch 'master' of https://github.com/Microsoft/CNTK into dongyu/BPTTWithContexts
...
Conflicts:
Source/ComputationNetworkLib/ComputationNode.h
2016-02-02 15:31:31 -08:00
21616b0019
make delimiter and decimal point customizable in UCIFastReader. This is a temp solution before the new reader is finished.
2016-02-02 14:09:15 -08:00
6d7c13a224
fixed CrossProcessMutex for Access Denied case
2016-02-02 13:42:55 -08:00
0dbd36e88f
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/devicePlacement
2016-02-02 13:14:59 -08:00
292ec03fc7
HTKMLFReader memory optimizations: Added string interning to the parsedpath type and changed logicalpath member type from utf16 to utf8
2016-02-02 13:13:46 -08:00
6f9a81ccee
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations
2016-02-02 13:08:58 -08:00
3fce4a8038
Merge branch 'master' of https://github.com/Microsoft/CNTK
2016-02-02 13:00:57 -08:00
1eeec90d65
documented a bug (inconsistent param name "needsGradient"), Issue #82
2016-02-02 13:00:34 -08:00
6668957d93
Integrate weixi/waynecoding into master
2016-02-02 11:04:02 -08:00
84c09e870e
add "using Base::IsPartOfLoop" statement in ComputationNode.h to make it accessible from inherited classes in gcc.
2016-02-02 09:57:22 -08:00
de415ccdc6
Merge branch 'master' of https://github.com/Microsoft/CNTK into dongyu/BPTTWithContexts
2016-02-02 09:48:59 -08:00
f90fa678a9
Fix batch normalization input
2016-02-02 09:45:22 -08:00
bfefbe2601
fixed error handling in CrossProcessMutex, now throws descriptive error instead of returning false in case of inrecoverable error (manually merged fseide/fixLock after git merge failed);
...
re-fixed ConvolutionNode::Load(), fix seemed to have gotten lost
2016-02-02 09:03:16 -08:00
8f8b5526a2
Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixMinor
2016-02-02 08:47:09 -08:00
b9d8961378
Fixing compiler issues + renaming
2016-02-02 14:34:21 +01:00
2251188bef
Fix ReaderLib.lib (and references)
2016-02-02 13:31:59 +01:00
93fb8d8283
Rename Source/Readers/{{New -> }ImageReader,Reader{ -> Lib}}
2016-02-02 13:24:37 +01:00
36771c9b37
Add {Debug,Release}_CpuOnly build configuration for NewImageReader and Readers
2016-02-02 13:24:37 +01:00
52d6444920
Add new image reader under new reader architecture
2016-02-02 11:30:15 +01:00
f628e52de2
Change number of bits used for utterance index in frameref to 11
2016-02-01 20:38:48 -08:00
86a356f13a
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/htkmlfreaderMemoryOptimizations
2016-02-01 17:55:26 -08:00
ba3b4b565e
throw error the delay node tries to access values out of boundary without carry over values when in a loop.
2016-02-01 16:42:25 -08:00
0fc4c0b3e8
Merge branch 'master' of https://github.com/Microsoft/CNTK into dongyu/BPTTWithContexts
2016-02-01 15:17:25 -08:00
7a9b0c038f
Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixMinor
2016-02-01 15:14:32 -08:00
31a212579e
Fix progress when epochsize is 0 and fix the git hash built into the exe
2016-02-01 12:56:47 -08:00
7c20c7225d
minor cleanup of Evaluate()'s out-of-date testing
2016-02-01 11:22:35 -08:00
e34b380407
In the DoWriteWordAndClassInfo command add code to generate intermediate folders for the vocabulary file.
...
Add "," and ";" as delimiters in UCIFastReader.
2016-02-01 11:14:36 -08:00
6e5bb7085e
Merge branch 'master' of https://github.com/Microsoft/CNTK into dongyu/BPTTWithContexts
2016-02-01 10:52:34 -08:00
d19da5c5e6
annotated a potential bug in SimpleOutputWriter::WriteOutput() (cf. Issue #73 )
2016-02-01 10:36:27 -08:00
bb85018744
in delay nodes, if boundary info is not available and past (or future) minibatches do not exist we use the first (for past value) and last (for future value) sample in the sequence as the past and future values. this is useful for things like truncated chunk or latency-controlled BLSTM setups.
2016-02-01 09:58:57 -08:00
734baf94c7
Fix #define
...
Activate LibSparsePCReader on Linux build
Make file name consistent with Windows
Activate LibSparsePCReader on Linux build
Make file name consistent with Windows
Activate LibSparsePCReader on Linux build
Make file name consistent with Windows
Fix reader on Linux
Activate LibSparsePCReader on Linux build
Make file name consistent with Windows
Activate LibSparsePCReader on Linux build
Make file name consistent with Windows
Activate LibSparsePCReader on Linux build
Make file name consistent with Windows
Activate LibSparsePCReader on Linux build
Make file name consistent with Windows
2016-02-01 17:42:42 +01:00
a7a14abfc2
Fix spacing
2016-02-01 17:42:40 +01:00
0980d03bda
Fix Windows build
2016-02-01 17:42:39 +01:00
0bbf3c4901
Fix reader on Linux
2016-02-01 17:42:38 +01:00
35fe2bffc1
Add Safeguard while loading a plugin
2016-02-01 17:42:36 +01:00
b0a6e00b8a
Use .cntk as default extension for CNTK configuration files
2016-02-01 10:19:48 +01:00
d7dbacd2fb
HTKMLFReader memory optimization: reduced the size of frameref struct
2016-01-31 23:22:53 -08:00
ee872eec54
HTKMLFReader memory optimization: Store phone boundaries in the utterance source only when lattices are there. This saves a huge amount of CPU RAM since the phone boundaries are stored for each frame.
2016-01-30 19:11:35 -08:00
1f2a8d85b2
Added strict error checking for free CUDA memory reporting during memory tracing
2016-01-30 17:52:30 -08:00
c4f9b262aa
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/devicePlacement
2016-01-30 16:34:34 -08:00
10fb06f2d0
ImageReader now understands imageLayout ids "cudnn" and "legacy" for mbFormat
2016-01-30 11:56:57 -08:00
a4f5b7a5d6
changed sample 02_Convolution to use imageLayout to enable cudnn use;
...
and fixed a refactoring bug in ImageParameter NDL definition
2016-01-30 11:56:55 -08:00
24411c2054
removed an unnecessary and incorrect consistency check
2016-01-30 11:56:54 -08:00
bee394dc0b
fixed an incorrect consistency check in TrainOrAdaptModel()
2016-01-30 11:56:54 -08:00
cd4d7e5b50
refined some error messages
2016-01-30 11:56:47 -08:00
e70b955230
Fixed hostname output in CudaCall and also made the CUDA device id reporting in CudaCall accurate by using cudaGetDevice
2016-01-30 11:10:57 -08:00
28d136b277
Made the single parameter ctor of Matrix types explicit to prevent their accidental usage for implicit conversion
2016-01-30 10:17:04 -08:00
2b08c6a402
Simplification of GPU device selection code with removal of the parallel error-prone GetBestDevice functionality in GPUMatrix. Now device selection is fully centralized and AUTOPLACEMATRIX functionality is removed
2016-01-30 03:03:12 -08:00
a5bcd71355
Merge branch 'master' of https://github.com/Microsoft/CNTK into jdroppo/FixEvalInputs
2016-01-29 17:43:07 -08:00
54037e5aea
Minor changes to the CUDA memomy allocation/free tracing
2016-01-29 13:40:50 -08:00
68000692a5
Merge pull request #51 from yzhang87/master
...
Fix the compile issue for Kaldi2Reader.
2016-01-29 10:22:51 -08:00
73faaa7687
Remove <Scc*> tags from .vcxproj files
2016-01-29 11:21:06 +01:00
68620bbb17
Add printMetadata flag to the dumpnode action. This command prints only values of the node.
2016-01-28 22:06:14 -08:00
91cf01e8d1
Fix CntkEval.dll Related Issues
...
Consumers of the CntkEval interface can pass in feature values
without label values. Most of the code was okay with this, but
the SimpleOutputWriter insisted on initializing both features
and labels. This caused GetMinibatchIntoNetwork function in
DataReaderHelpers.h to fail, when the layout of the label node
was incorrect.
I've changed the SimpleOutputWriter to only use input nodes
that are ancestors of the desired output nodes. This fixes the
issue for CntkEval usage.
2016-01-28 17:26:13 -08:00
b8badf66b1
let->const auto
2016-01-28 16:16:22 -08:00
959d3dd5df
Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixRowSlice
2016-01-28 16:15:30 -08:00
f1b1fd73f8
Fix the compile issue for Kaldi2Reader.
2016-01-28 16:41:09 -05:00
2b2d5d6b21
Add Windows flavor of file sync
2016-01-28 11:39:18 -08:00
4c7919a71f
add fsync to fflushordie
2016-01-28 10:20:02 -08:00
d09a3babda
Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/fixRowSlice
2016-01-28 09:49:03 -08:00
c7c5403994
tensor support for RowSlice operation;
...
completed CudaErrString<cublasStatus_t>()
2016-01-28 09:40:57 -08:00
3224f02626
#5 change hardcode environment to system envrioments
2016-01-28 22:23:18 +08:00
203aa2115c
NoGPU.cpp: fix RuntimeError() messages
2016-01-28 14:21:52 +01:00
b6995dd80b
Fix Windows {Debug,Release}_CpuOnly builds (without 1-bit SGD)
2016-01-28 12:07:24 +01:00
37fb8df40a
(some reordering of methods in ReshapingNodes.h--CopyTo, Load, Save, forward, backward, Validate, no code change)
2016-01-27 09:42:16 -08:00
4fe5fd7251
Modified projects and solution for a new CPUONLY configuration (Debug_CpuOnly and Release_CpuOnly)
...
Added missing CRT_SECURE_NO_WARNINGS to latticeNoGpu.cpp
Added missing reference to Matrix<double> (causing linking issues in Release_CpuOnly)
2016-01-27 16:16:37 +01:00
7c280331bf
Fixed https://github.com/Microsoft/CNTK/issues/15
2016-01-25 20:00:53 +01:00
53de59c1c7
Added CLI Wrapper to native Evaluation Model
...
Added C# Client for CLI wrapper
2016-01-25 15:08:54 +01:00
38778e103c
Merge branch 'master' of https://github.com/Microsoft/CNTK
2016-01-25 13:56:44 +01:00
e8bda6883b
Merge branch 'master' of https://github.com/Microsoft/CNTK
2016-01-25 10:26:16 +01:00
2462647f62
White-space (remove CRs that crept in)
2016-01-25 10:24:10 +01:00
76aef066cc
Fixes to broken / missing paths in *.sln and *.vcxproj*
...
(Via check check-git-vs-vs.ps1)
2016-01-25 10:12:46 +01:00
cb0b09fbd8
Source/1BitSGD: change to new hash again, f785679a6b
2016-01-25 09:18:42 +01:00
26911b7471
made gcc happy
2016-01-23 21:07:36 -08:00
2f640310b6
rewrote RowStackNode using the tensor lib
2016-01-23 20:57:14 -08:00
161e6fb513
Merge branch 'amitaga/htkmlfreaderFix'
2016-01-23 10:53:57 -08:00
e19cc4cf09
Update location for Source/1BitSGD
2016-01-23 07:23:12 +01:00
af97079a58
removed #if-0'ed out code
2016-01-22 17:34:41 -08:00
6dff1a8ea0
deleted 5 unused methods
2016-01-22 17:16:07 -08:00
d48ec6f66e
sorting and editing of ComputationNode<ElemType>, no code change
2016-01-22 17:07:20 -08:00
b29b352845
accidentally removed SequenceWithSoftmaxNode, now back;
...
sorting of stuff in ComputationNodeBase, no code change
2016-01-22 16:45:33 -08:00
b0bbc0c949
marked CRFNode as #ifdef COMING_SOON
2016-01-22 15:40:42 -08:00
198d64dd64
regularized the naming convention in SimpleNetworkBuilder.cpp/h, and added comments from Kaisheng on network kinds
2016-01-22 15:18:05 -08:00
55b3ae946f
marked the following nodes as COMING_SOON: TransposeNode, GMMLogLikelihoodNode, SequenceDecoderNode. These will be brought back once we have defined tests for them and made necessary updates;
...
moved DiagonalNode to ReshapingNodes.h;
renamed DeprecatedReshapeNode to LegacyReshapeNode
2016-01-22 14:46:30 -08:00
7f07161ccd
removed trailing spaces
2016-01-22 14:06:17 -08:00
a4cf44544a
normalized inconsistent commenting style from "//text" to "// text"
2016-01-22 13:58:47 -08:00
f79fe393b7
Fixed a linux only reader bug where forward slashes in pageFilePath were being replaced with backslshes even on Linux
2016-01-22 13:50:57 -08:00
27641d86d0
normalized comment separator from /// to //
2016-01-22 13:47:40 -08:00
03a4fcb295
renamed TrainingCriterionNodes.h to TrainingNodes.h, and moved DropoutNode and BatchNormalizationNode there;
...
renamed EvaluationCriterionNodes.h to EvaluationNodes.h, and moved SequenceDecoderNode there
2016-01-22 13:45:14 -08:00
6c74cc5aa5
removed some unused methods and skipPairNetwork arg to EnumerateNodes from ComputationNote
2016-01-22 11:46:23 -08:00
abd520633a
deleted the following nodes, as they are either redundant or currently not implemented correctly (but may come back in the future): StrideTimesNode, ParallelNode, LSTMNode, BatchModeNode, TimeReverseNode;
...
also sorted some of the lookup-table like functions (matching node names) alphabetically
2016-01-22 11:25:52 -08:00
f4197414de
renamed EsotericNodes.h to SpecialPurposeNodes.h, and CompositeComputationNodes.h to PreComputeNodes.h;
...
moved GMMLogLikelihoodNode and SequenceWithSoftmaxNode to SpecialPurposeNodes.h;
also moved TimeReverseNode.h there (which will be deleted soon);
replaced the min/max macro use with std::min/max;
added all frequently used C/C++ headers into all stdafx.h
2016-01-22 10:45:53 -08:00
7b6708cb9c
removed condtional compilations for tensor view
2016-01-22 09:57:10 -08:00
6d31cda88a
removed two more Multinetworks configs from SimpleNetworkBuilder
2016-01-22 09:42:57 -08:00
a4a20183aa
made everything build again after deleting the unused MultiNetworks functions, incl. deleting related entries in SimpleNetworkBuilder;
...
deleted PairNetworkNode;
renamed EsotericActions.cpp to SpecialPurposeActions.cpp
2016-01-22 09:13:52 -08:00
18b7b36cca
removed MultiNetworksSGD.h and -Evaluator.h, not yet made it compile
2016-01-22 08:28:57 -08:00
a0a32466a9
White-space
2016-01-22 10:15:34 +01:00
6f59eb1d22
Formatting
2016-01-22 09:23:03 +01:00
50c0e8e345
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/memoryFootprint
2016-01-21 22:00:54 -08:00
59287d0b09
fix bug in imageLayout parameter and defaultHiddenActivity parameter in NDL usage prompt
2016-01-21 18:43:35 -08:00
8cf0b86b54
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/memoryFootprint
2016-01-21 18:33:40 -08:00
890dda9fde
include imageLayout optional parameter in the usage prompt of Convolution, maxpooling, and avgPooling. in NDL
2016-01-21 18:28:44 -08:00
083dbacc3f
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/memoryFootprint
2016-01-21 18:14:05 -08:00
6137d501ee
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/preComputeProgressTracing
2016-01-21 18:05:39 -08:00
fd90701166
make gcc happy
2016-01-21 16:29:44 -08:00
c177d3abac
Added progress tracing for PreCompute phase - currently just printing 0.0 instead of actual timeout which is sufficient for the current purpose
2016-01-21 16:26:51 -08:00
04f3089554
removed a now invalid consistency check
2016-01-21 16:14:54 -08:00
416ead4138
fixed verification of size of loaded learned parameter values: now works with non-matrices, too
2016-01-21 16:08:09 -08:00
14fee02e39
modified ComputationNode's matrix storage back to match Matrix dimensions if no minibatch, to ensure compatibility with sparse matrices
2016-01-21 15:00:12 -08:00
c9530e85a1
updated PerformSVDecomposition() to use ValueAsMatrix()
2016-01-21 12:59:47 -08:00
ce0c087936
some consistency-check fixes related to SetDims() vs. MBLayout
2016-01-21 12:52:30 -08:00
8e7b19b566
bug fix: LazyZeroGradient() must not use m_value dimensions for resizing m_gradient, because m_value may have been lost already due to memshare;
...
bug fix: DelayedValueBase should not call SetDims(., false) because upon re-load (rollback) there is already an MB Layout, which will cause a consistency check to fail
2016-01-21 11:57:03 -08:00
2c9f56a205
Added ability to ignore cudaFree return code in the tracing allocator since free is sometimes called during process exit when cudaFree calls would fail.
2016-01-21 11:19:53 -08:00
4691595f0f
bug fix, ClassBasedCrossEntropyWithSoftmax now working
2016-01-21 10:31:55 -08:00
797a9e0a13
Fixed linux build
2016-01-21 08:21:05 -08:00
5bbc37f508
bug fix: TableLookupNode still used one wrong matrix interpretation
2016-01-21 07:56:47 -08:00
e389459ca6
LearnableParameter::InitRandom() now mimics the Matrix initialization behavior;
...
bug fix: LookupTableNode used GetAsMatrixNumRows() on input[1] which is a minibatch;
bug fix: Image/QuickE2E network definition updated to drop the now unnecessary extra column dimension;
bug fix: TensorShape::IsDense() should not require m_offset to be 0 (column slices are perfectly fine)
2016-01-20 23:28:33 -08:00
d44c9efdc9
made gcc happy
2016-01-20 21:44:33 -08:00
f1a342c6a4
Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/tensors
2016-01-20 21:36:09 -08:00
342b1ba460
changed the mapping of tensors onto the Matrix storage objects: In case of no MBLayout, the Matrix objects now always have column dimension 1. Actual matrices (as used by TimesNode) are now 2D tensors. As part of this, GetNumCols()/-Rows() no longer exist, but were split into GetSampleMatrixCols()/-Rows() (minibatch interpretation) and GetAsMatrixCols()/-Rows() (2D matrix interpretation)
2016-01-20 21:33:38 -08:00
ffb930a18d
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/memoryFootprint
2016-01-20 17:24:17 -08:00
283908359f
Added free memory reporting as part og GPU memory allocation/free traces and some refactoring
2016-01-20 17:20:12 -08:00
31a164602c
Fixed topk memory leak.
2016-01-20 11:07:16 -08:00
fa939cc641
CR Fix - Switched copyright headers to MIT license
...
CR Fix - Removed EvalDll.lib linking
2016-01-20 14:45:27 +01:00
cc21bb7b87
Fixed CPUONLY build
2016-01-19 23:29:28 -08:00
cf14913860
Added CUDA memory allocation/free tracing facility to GPUMatrx and GPUSparseMatrix
2016-01-19 22:19:18 -08:00
01bdefe2d5
Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/tensors
2016-01-19 09:08:08 -08:00
678176562d
merged fseide/tensors and master after github move
2016-01-19 08:49:26 -08:00
054ac358e7
Merge branch 'master' of https://github.com/Microsoft/CNTK
2016-01-19 11:51:43 +01:00
d582eb2701
Cleanup and commented for CR
2016-01-19 11:47:16 +01:00
1ae3404a74
moved ParallelNode to EsotericNodes.h (it is redundant with RowStackNode)
2016-01-18 16:50:09 -08:00
78a7edeb36
BestGpu::Init() no longer throws when run under remote access
2016-01-18 16:40:31 -08:00
742e239392
Minor: Pending CR feedback - fixed some typos in comments
2016-01-18 15:03:40 -08:00
250321031b
Added Managed wrapper to Eval interface
...
Added CS Client for managed wrapper
2016-01-18 14:47:32 +01:00
f1c3232024
Missing tabs-to-spaces conversion
2016-01-18 09:40:10 +01:00
6c5aa5809e
Add submodule for 1bit-SGD
...
This is available under a different license,
see https://github.com/Microsoft/CNTK-1bit-SGD for more information.
2016-01-18 09:36:24 +01:00
dcfc0e9679
License change
2016-01-18 09:36:17 +01:00
2f39fe3f01
Re-format code using clang-format (plus some post-processing)
2016-01-18 09:36:14 +01:00
1bcefe770a
.gitattributes: set proper configuration for text files (and normalize files)
2016-01-18 09:35:54 +01:00
146de2650d
Merge branch 'master' of https://git01.codeplex.com/cntk into amitaga/memshare
2016-01-17 22:00:31 -08:00
4f2a597591
Bugfix: Enabled output value memory sharing for non-BS CNTK configs too
2016-01-17 21:44:19 -08:00
fa4be28337
Merge branch 'master' of https://git.codeplex.com/cntk into fseide/tensors
2016-01-17 17:44:14 -08:00
97fadca198
added some consistency checks for m_numCols in prep for removing it
2016-01-17 17:42:35 -08:00
3d87c905fd
Added a temporary comment about about the MPI v7 requirement for MPI_Iallreduce call on Windows. This comment will be removed in a few weeks.
2016-01-17 17:11:28 -08:00
e37fede6d7
updated the link in MPIWrapper.h to the required MPI SDK installer;
...
ComputationNode::GetNumRows() now returns its value directly from m_sampleLayout, the m_numRows member has been deleted
2016-01-17 16:20:01 -08:00
582b03d45f
Some minor node value matrix memory sharing fixes
2016-01-16 18:13:03 -08:00
1805830c6a
made gcc happy (ptrdiff_t vs. int)
2016-01-16 00:07:38 -08:00
b933d83f3f
Merge branch 'master' of https://git.codeplex.com/cntk into fseide/tensors
2016-01-15 23:53:06 -08:00
267be9eedd
completed ShiftNode::BackpropTo()
2016-01-15 23:50:06 -08:00
c186d49ff0
added specializations to linear unary non-reducing ops: copy and add ops now use existing CUDA functions;
...
ShiftNode backprop functional for non-packed case/no boundary state
2016-01-15 17:08:26 -08:00
4ac49b0cf2
ShiftNode::ForwardProp() implemented
2016-01-15 11:21:25 -08:00
8ae5176a21
Merge branch 'master' of https://git01.codeplex.com/cntk into amitaga/separate1bitDataParallelSGD
2016-01-15 09:40:43 -08:00
e9a91a8d38
Merge branch 'master' of https://git01.codeplex.com/cntk into bmitra/Changes
2016-01-15 03:30:27 -08:00
3ce17cbc94
Bug fix for _tensorShuffleScaleAndAddRowSparse kernel.
2016-01-15 03:30:15 -08:00
e32329a85c
Merge branch 'master' of https://git01.codeplex.com/cntk into amitaga/separate1bitDataParallelSGD
2016-01-15 00:08:17 -08:00
0f2786c4f1
Added support for 1BitSGD build flavor to build-and-test script. Also fixed a CPUONLY build error
2016-01-14 10:58:31 -08:00
dd858fa8c5
Minor GPU sparse changes.
2016-01-14 09:46:31 -08:00
b3f7361fe5
TrainOrAdaptModel, when doing regularized adaptation, now properly INvalidateCompiledNetwork() before calling CompileNetwork() again.
2016-01-14 09:00:04 -08:00
f85ef912a4
FromEvalOrder() warning message had possible nullptr dereference, fixed.
2016-01-14 09:00:03 -08:00
ada6b579c5
Minor sparse changes.
2016-01-14 07:41:43 -08:00
8d646e1c2a
removing zombie vcxproj files
2016-01-14 11:12:20 +01:00
9e9cab049d
removing files and folders as announced (examples and readers)
2016-01-14 09:06:40 +01:00
fc61bb187a
Removed a few unreferencep source files
2016-01-14 09:03:10 +01:00
18528f15b4
Merge branch 'master' of https://git01.codeplex.com/cntk into amitaga/separate1bitDataParallelSGD
2016-01-13 22:43:27 -08:00
40cbeac00b
more strict control of compiled/non-compiled state of networks in MEL. All editing operations invalidate compilation, and Saving the model will validate the model if changed (if not it was validated during Load)
2016-01-13 19:32:49 -08:00
c89b6320dc
Merge branch 'master' of https://git.codeplex.com/cntk into fseide/tensors
2016-01-13 18:56:33 -08:00
b3cfd6e45d
bug fix in TensorSliceWithMBLayoutFor() handling of m_timeOffset for entire batch
2016-01-13 18:54:01 -08:00
31b0e7098e
added a new file-format option to the MEL LoadModel command in order to load older CNTK files that had borked tensor dimensions;
...
FrameRange now allows time ranges
2016-01-13 18:42:17 -08:00
de5be29239
Minor performance upgrade on row slicing to avoid GPU mem copy.
2016-01-13 15:45:20 -08:00
f835efd052
Fix for multi GPU to share all parameters required to adjust learning rate.
2016-01-13 15:45:19 -08:00
d39d87f03b
Fixed a bug where the m_elemSizeAllocated was used instead of m_nz
2016-01-13 15:45:18 -08:00
b66ea69666
Remove 1bit SGD implementation headers references from the CNTK project files
2016-01-13 14:46:33 -08:00
55ccb09681
Merge remote-tracking branch 'origin/master' into myseq
2016-01-13 14:16:12 -08:00
19a9895d1a
Print SeqGammar related parameters for better logging.
2016-01-13 12:25:20 -08:00
07de5fa533
Moved the 1bit SGD specific code to separate folder and changed Windows project files to pivot the 1bit support off of an environment variable
2016-01-13 11:20:30 -08:00
ba61abd79e
Disable popups on Windows
2016-01-13 10:11:28 -08:00
70cea38b08
Merge branch 'master' of https://git01.codeplex.com/cntk into amitaga/separate1bitDataParallelSGD
2016-01-13 09:58:40 -08:00
270726e324
Make SequenceGammar calculation parameters configurable.
2016-01-12 23:58:28 -08:00
f0655f04d9
(make gcc happy)
2016-01-12 18:12:11 -08:00
c45401fbd0
Add Nesterov's momentum.
...
to use NAG, simply add
useNAG=true
2016-01-12 17:08:46 -08:00
b61d6704ec
Merge remote-tracking branch 'origin/master' into myseq
...
Conflicts:
Source/CNTK/ModelEditLanguage.cpp
Source/Common/Include/Sequences.h
Source/ComputationNetworkLib/ComputationNetwork.h
Source/ComputationNetworkLib/ComputationNetworkEditing.cpp
Source/ComputationNetworkLib/ConvolutionalNodes.h
2016-01-12 15:57:54 -08:00
5bb9fbf6e6
Fix a bug pointed by Alexey. Thanks!
2016-01-12 15:50:02 -08:00
914ac61c96
Fix an inconsistency after merge with master.
2016-01-12 15:27:32 -08:00
92e8a4d136
Added BN eval mode to MEL. Updated samples.
2016-01-12 14:47:32 -08:00
9e25b7e61a
Removed Resize from BN code. Updated samples.
2016-01-12 14:02:51 -08:00
cc2a836c85
Updated samples, added ResNet-50.
2016-01-12 14:02:43 -08:00
f52e80cf8f
Added CMA to BN node, updated samples.
2016-01-12 14:02:37 -08:00
b31b921a35
Linux build fix.
2016-01-12 03:39:31 -08:00
052544d177
Linux build fix.
2016-01-12 03:23:30 -08:00
48133a8b63
Linux build fix.
2016-01-12 03:12:35 -08:00
f76412385d
Bug workaround: The m_columnsValidityMask matrix in MBLayout type was being default initialized resulting in incorrectly selecting a bad GPU device.
2016-01-11 22:45:17 -08:00
25fd18bf1f
Fix an error in SequenceWithSoftmaxNode::RequestMatricesBeforeForwardProp
2016-01-11 17:26:52 -08:00
314f057086
Merge branch 'master' into myseq
...
Conflicts:
Source/ComputationNetworkLib/ComputationNode.h
Source/ComputationNetworkLib/InputAndParamNodes.h
Source/ComputationNetworkLib/TrainingCriterionNodes.h
2016-01-11 17:07:56 -08:00
2d7b74e825
(Fix a bug in MarkValueSharable)
2016-01-11 16:54:08 -08:00
8aa59f7009
(further remove MarkValueNotSharable out of constructor)
2016-01-11 16:51:45 -08:00
60989d7acb
Move MarkValueNonsharable out of constuctors
...
(make gcc happy)
2016-01-11 16:51:28 -08:00
6777bbe0b7
Display CUB and CUDNN paths (if defined) in BuildInfo
...
Print BuildInfo at the very begining of the program. convenient for checking
build type.
2016-01-11 16:49:00 -08:00
b54cfccc37
Add an alternate option "numSubminibatches" for users to indicate how to split
...
minibatches into subminibatches.
2016-01-11 16:48:50 -08:00
2f51fb24b6
Fix a bug in MarkValueSharableNode
2016-01-11 16:48:40 -08:00
0b8e30ea22
Revise the condition of ReleaseMatricesAfterForwardProp: only ValueSharable
...
nodes can be released after forwardprop
2016-01-11 16:48:29 -08:00
4b1f8006b3
Fix MarkValueNotSharableNodes
2016-01-11 16:48:13 -08:00
485a7b8fe6
Revise the implementation of valueNotSharableNode. More to be revised.
2016-01-11 16:47:56 -08:00
edbb47dc79
Replace CreateMatrixIfNull by MarkValueNonsharable()
...
In the compiling the stage, we will mark nodes as nonsharable whose descendents are all learnable parameters.
2016-01-11 16:47:46 -08:00
569a4d6c21
Add support for revising batch normalization property in MEL.
...
Now a BatchNormalization node's eval mode can be modified by
SetProperty(BNnode, batchNormEvalMode, true);
or by
SetPropertyForSubTree(rootNode, batchNormEvalMode, true);
in which all the BN nodes under rootNode will be changed.
2016-01-11 16:32:19 -08:00
bc3b96e79a
Merge branch 'myseq' of //vilfblgpu011/d/users/erw/src/cntk into myseq
2016-01-11 12:29:22 -08:00
7c434383d7
Reverting a minor change.
2016-01-11 10:05:16 -08:00
3c4cddad19
Merge branch 'master' of https://git01.codeplex.com/cntk into bmitra/Changes
2016-01-11 07:31:31 -08:00
54fffbd367
Minor fixes.
2016-01-11 07:31:17 -08:00
11ea65bc20
Minor fix to transpose function.
2016-01-11 01:57:56 -08:00
05e0262bf1
Bug fix for ConvertDBN command
2016-01-09 23:17:51 -08:00
c41eafa4f8
Move all data member in MBLayout to CPU
2016-01-09 23:16:45 -08:00
4d9304c49e
removed multi-dim feature from ShiftNode, causes too many inconsistencies
2016-01-09 21:34:44 -08:00
08e7d59059
moved all actions (DoXXX()) from CNTK.cpp to ActionsLib (no code change);
...
reenabled tensor lib (undid accidental commit)
2016-01-09 18:13:31 -08:00
69b547728a
(made sure non-tensor version still compiles)
2016-01-09 17:03:20 -08:00
0826c1ccbb
moved non-tensor versions of PlusNode, MinusNode, and ElementTimesNode to EsoreticNodes.h (no code change)
2016-01-09 16:58:26 -08:00
a0fc021da0
DataTensorFor() refactored further
2016-01-09 16:51:45 -08:00
c886f32d6d
factored out new function TensorSliceWithMBLayoutFor() from DataTensorFor(), for use by ShiftNode
2016-01-09 16:30:25 -08:00
571c192f6d
Remove reshapeInputToRowSize in SparsePCReader.
2016-01-09 03:02:21 -08:00
8ed236d87a
Merge branch 'master' of https://git01.codeplex.com/cntk into bmitra/Changes
2016-01-09 02:42:11 -08:00
84afb7dd9b
Changes to text convolution.
2016-01-09 02:41:47 -08:00
Frank Seide
thhoens
Amit Agarwal
Project Philly
Mark Hillebrand
Eldar Akchurin
Philipp Kranen
U-FAREAST\fseide
Mark Hillebrand
Dong Yu
Wayne Xiong
Chris Basoglu
Clemens Marschner
Jasha Droppo
Dong Yu
vadimma
yzhang87
Marko Radmilac
Qiwei ye
Gaizka Navarro
pannous
Alexey Kamenev
bmitra
Yongqiang Wang