a5d8a61861
merged from fseide/initial_state
2017-02-16 17:13:33 -08:00
2eaef84d7b
update char_rnn_test.py to match char_rnn.py
...
train_and_eval_char_rnn() now takes both epochs and total minibatch limit.
2017-02-16 16:59:40 -08:00
61baf021cc
word_rnn
2017-02-16 21:26:30 +01:00
df0d911b85
Rebase CTC changes
2017-02-16 09:32:55 -08:00
0bf2f4636c
Tests/Install/windows: make pre-filling InstallCache optional
2017-02-16 16:09:44 +01:00
a63c43d153
Tests/EndToEndTests/ParallelTraining/AsynchronousSGD: run against single GPU
2017-02-16 16:09:44 +01:00
3e87868390
Tests/EndToEndTests/CNTKv2Python/Examples/AlexNet_ImageNet_Distributed_test.py: decrease MB size for test
...
Occasionally (< 1 % of runs) the test hits OOM on some of our test machines.
2017-02-16 16:09:44 +01:00
76a1aa4e84
add C# tests on GPU
2017-02-16 09:19:34 +01:00
ba3f24b74e
Exposing EditDistanceError node in BrainScript
2017-02-15 23:20:09 -08:00
d06c4ca519
Rename CTCNode to ForwardBackwardNode
2017-02-15 23:14:29 -08:00
f37bc958e2
merged from master
2017-02-15 15:45:32 -08:00
8e2f5efcf1
address review feedback, iteration 2
2017-02-14 13:08:37 -08:00
3bebb0d4f4
Tests/EndToEndTests/CNTKv2Python/Examples/conftest.py: remove Jupyter Notebook fixture
2017-02-14 17:50:35 +01:00
7d0febedba
Splitted Python e2e test into examples and tutorials
2017-02-14 12:58:20 +01:00
b92d09e583
updated baseline
2017-02-14 10:21:30 +01:00
d4e3fcdd98
fixed DFS bug, added test
2017-02-14 09:46:24 +01:00
261ee40809
merged from master
2017-02-13 21:14:29 -08:00
5f350c2c0f
address review comments
2017-02-13 16:05:35 -08:00
7bd974c894
Fixing error when the same image from zip refereced several times in the mapping file
2017-02-10 16:52:13 +01:00
0673df13da
Merge branch 'master' of https://github.com/Microsoft/cntk into fseide/layers-lib-breaking
2017-02-09 19:52:41 -08:00
752c59e02f
Updated docs, test and nb spell mistakes
2017-02-09 19:35:30 -08:00
fccd6a48bd
Merge branch 'master' of https://github.com/Microsoft/CNTK into guschmue/tut-106b
2017-02-09 18:11:29 -08:00
cee8cd1601
Updated the branch to current master and update CNTK 106B
2017-02-09 18:08:57 -08:00
adf0a9238a
changed signature of splice() from list to *args, to make it consistent with all other ops;
...
renamed cntk.ops.reduce_log_sum() to reduce_log_add_exp(), to make it consistent with log_add_exp();
associative binary ops like plus() now accept >2 arguments;
new ops element_max(), element_min();
softmax() now takes axis parameter;
constant() now takes dtype parameter argument
2017-02-09 17:06:09 -08:00
19d2e5f7ea
changed all imports of blocks and models to layers
2017-02-09 15:41:12 -08:00
74fa7f440f
CNTK v2 library: Implement a true no-op operation for an efficient Alias creation that does not use data pass through
2017-02-09 13:34:22 -08:00
a54ba79054
Integrate pkranen/transferLearning into master
2017-02-09 10:57:58 -08:00
5f938516d0
Integrate thilow/SimpleExampleSSM into master
2017-02-09 09:29:50 -08:00
4214003a38
Fixed e2e test for transfer learning example
2017-02-09 15:24:49 +01:00
b7d4945a8e
Integrate t-ivrodr/v2tests_to_boost_squashed into master
2017-02-09 05:24:10 -08:00
2c1e71244f
Tests/Install/linux/prep-run-test.sh: choose specific kernel when running Jupyter NBs
2017-02-09 12:08:24 +01:00
98c2fbdcd1
Merge remote-tracking branch 'origin/master' into thilow/SimpleExampleSSM
2017-02-09 12:01:11 +01:00
962593932e
Merge branch 'master' into t-ivrodr/v2tests_to_boost_squashed
...
# Conflicts:
# Tests/UnitTests/V2LibraryTests/LearnerTests.cpp
2017-02-09 11:27:29 +01:00
236fa8ac9f
Tests/EndToEndTests/UnitTests/MathTests/baseline.txt: missing baseline update
2017-02-09 10:53:43 +01:00
04db23b5f8
Add Adam Learner as the original paper
2017-02-09 08:34:47 +00:00
6829622172
Disable test for VGG16. Jenkins ran out of memory. Will enable it once we finish memory optimization.
2017-02-08 17:05:47 -08:00
f9052ea5f3
Revision due to changes in train_session. Also rewrote tests due to failure caused by single thread running of AlexNet test and VGG test.
...
Bug fix in ReshapingNodes.h.
2017-02-08 15:46:45 -08:00
74ec3ab2d3
Add test for VGG16.
2017-02-08 15:46:44 -08:00
50e8e10406
Merge remote-tracking branch 'origin/master' into thilow/SimpleExampleSSM
2017-02-08 16:15:44 +01:00
57b7587917
Converting current v2 library unittests to boost framework.
...
Some of them were joined in a single project of e2e tests.
2017-02-08 15:22:49 +01:00
1ae666d431
Added extended transfer learning example
2017-02-07 23:37:44 +01:00
1ab395aa76
Added Flowers data set and scripts
2017-02-07 23:37:37 +01:00
86a20080b1
Integrate eldak/enhancingTrainingSession into master
2017-02-07 13:09:28 -08:00
62cf5c7911
Integrate zhouwang/refactor-props into master
2017-02-07 12:25:31 -08:00
7f9850f4a5
Integrate alrezni/improve_ctf into master
2017-02-07 11:17:50 -08:00
ad78b15f29
Enhancing the training session with CV functionality
2017-02-07 19:26:50 +01:00
42cb83940b
Tests/Install/windows/.dockerignore: normalize LEs
2017-02-07 18:41:29 +01:00
13cf9de3e1
further simplify project settings
2017-02-07 18:29:24 +01:00
fe765f352a
create cntk.common.props to share common project property between C++ and C#; refactor related project files to use cntk.common.props
2017-02-07 18:29:24 +01:00
64a1ee8dd1
Integrate mahilleb/WinOobeTest into master
2017-02-07 08:32:43 -08:00
f5df13408a
CTF: fix default randomization, missing newlines
...
Support chunk randomization window specified in the number of chunks (not
samples). Use chunk-based window by default in CTF Reader (when no
randomization window is explicitly specified). In V2, use chuck-based
randomizaton by default, with a default window size = 128 chunks. Handle
trailing input lines not ending with newlines (print a warning to stderr,
but do not throw, as long as the line ends with a valid value).
2017-02-07 17:30:46 +01:00
dd49de6e62
Tests/Install/windows: tune slightly
2017-02-07 16:06:37 +01:00
a3380bcd61
Tests\Install\windows: 1st version of Windows OOBE test added
2017-02-06 17:46:34 +01:00
8176e0fccf
add Function::FindByName(), Function::FindAllWithName; Add tests;
2017-02-06 11:42:45 +01:00
4f4b5fb52f
CNTK v2 library: Fix variable_value_to_seq conversion from Value to numpy array list, for cases where the Value objects contains sequence segments instead of full sequences. Also added a regression test for the char RNN example.
2017-02-03 18:18:35 -08:00
fbb53d2d92
Bundle all fseide fixes for denom sharing in BN
...
Collect all of Frank's changes related to fixing the shared denominator
in Batch Normalization node in a single commit. Patch up BS/NDL config
builders and model loaders to add an extra input for 'running mean
sample counter' on the fly. Same for loading V2 model the only define 5
inputs for BN functions.
2017-02-03 16:31:05 +01:00
62654a1594
Integrate eldak/fixingInferedOutputs into master
2017-02-02 07:14:23 -08:00
8bd31d70b3
Integrate zhouwang/fix-v2evaltests-optional2 into master
2017-02-02 07:08:39 -08:00
e4597c9461
change reference type from project reference to file refernce to make msbuild happy; add prect dependency
2017-02-02 14:59:35 +01:00
5d2495b46f
Fixing inferrence of outputs to allow crossing of dll boundary
2017-02-02 14:47:52 +01:00
566355231c
CNTK v2 library: Add support for precompute nodes in LoadLegacyModel
2017-02-02 01:29:19 -08:00
e0feaefcf0
CNTK v2 library: 1. Allow parameter and constant objects to directly use a
...
preallocated NDArrayView instead of always creating a clone 2. Fix an
issue with loading of legacy convolution models where the parameter tensor
is shaped as 2D instead of the real convolution map shape
2017-02-01 21:15:04 -08:00
7ea2ff157d
CNTK_207_Training_with_Sampled_Softmax: adding test. Adding device-awareness in notebook.
2017-02-01 15:39:07 +01:00
27b18215f7
Integrate eldak/increasingTolerance into master
2017-02-01 05:23:18 -08:00
85af41723e
Increasing tolerance
2017-02-01 11:40:38 +01:00
1e6febcfa7
add CreateBatch/Sequence/BatchOfSequence to C++
...
add tests
Use NDArraryView to avoid extra copy
2017-02-01 09:22:54 +01:00
5170d4a2ce
make CNTKLibraryCSEvalExampletest project optional for build
...
use the same conditional compilation pattern as in C++
2017-01-31 13:38:25 +01:00
5d1a014956
merged from master
2017-01-30 16:49:10 -08:00
cd99733472
Merge branch 'master' of https://github.com/Microsoft/CNTK into sayanpa/gantut
2017-01-30 13:43:50 -08:00
aa884114cd
added cntk-106b
2017-01-30 13:27:56 -08:00
feb94e3354
GAN test threshold and tollerance update for GPU
2017-01-30 09:24:59 -08:00
cfcf73cd22
Integrate eldak/distributedTesting into master
2017-01-30 09:16:15 -08:00
717f198400
Adding more distributed tests for python
2017-01-30 15:29:45 +01:00
a557e9ef73
fixed tolerance for flaky GAN test
2017-01-30 14:11:25 +01:00
0fcff03775
sequence_to_sequence_test.py updated w.r.t. Layers lib
2017-01-27 19:21:48 -08:00
a56fdd518f
updating threshold for basic gan test for CPU
2017-01-27 19:05:30 -08:00
2126d008b1
Merge branch 'master' of https://github.com/Microsoft/CNTK into sayanpa/gantut
2017-01-27 17:38:27 -08:00
ff9337a812
Added tollernce per device to GAN tutorial
2017-01-27 17:38:11 -08:00
1ac5d3be26
Changing basic GAN test target for CPU and tollerance
2017-01-27 15:50:39 -08:00
7043deac0f
GAN tutorial test threshold updated
2017-01-27 14:08:21 -08:00
d628aa377a
Fix bug that fails test due to relative path import.
2017-01-27 12:45:20 -08:00
b3da8a211d
more test updates w.r.t. Layers lib
2017-01-27 11:23:36 -08:00
2e4b4b559c
Merge branch 'master' of https://github.com/Microsoft/CNTK into sayanpa/gantut
2017-01-27 11:15:43 -08:00
d6559727af
updated more tests w.r.t. Layers
2017-01-27 10:40:01 -08:00
e13224113b
Update based on CR comments.
2017-01-27 10:30:59 -08:00
c78b254efe
Code revision based on CR comments.
2017-01-27 10:30:59 -08:00
296bdfc841
Adding a test for AlexNet training. Revised the CIFAR-10 tests so that the load data correctly.
2017-01-27 10:30:59 -08:00
faa973b8f5
Added GAN tutorial and fixed test threshold
2017-01-26 18:40:28 -08:00
9964293f8f
Added GAN tutorial with CR feedback
2017-01-26 17:57:57 -08:00
8068dcc056
Merge branch 'master' of https://github.com/Microsoft/CNTK into sayanpa/gantut
2017-01-26 16:13:27 -08:00
48e85566ca
Add python unit tests for edit_distance_error
2017-01-26 09:19:23 -08:00
593327adc0
EditDistanceError: Change samplesToIgnore to vec of size_t
2017-01-26 09:19:23 -08:00
2de82e84a7
Integrate mahilleb/TestMoves into master
2017-01-26 06:16:45 -08:00
8a180ba142
Making owner in output variables ref counted
2017-01-26 12:46:45 +01:00
05db810634
Swinging some of the parallelization over to HTKDeserializers
2017-01-26 09:53:40 +01:00
f94ebcf4f6
Adding basic GAN tutorial
2017-01-25 16:07:19 -08:00
f8bf543cad
merged from master
2017-01-25 13:57:39 -08:00
520b46d523
remove unnecessary TODOs
2017-01-24 17:20:17 +01:00
af8631ef9c
changes for pascal card
2017-01-24 12:01:41 +01:00
24d85865e7
Fixed BS deconv test after config change
2017-01-24 09:52:37 +01:00
e57a497723
Added test for Python deconv example
2017-01-24 09:52:37 +01:00
16daa4acb5
Speed up TimesNodeBase for sparse by avoid unroll
...
This change improves speed of CrossEntropyWithSoftmax/ClassificationError with sparse label. Example/LanguageUnderstanding/ATIS/Python/LanguageUnderstanding.py got 4x speed up.
2017-01-23 18:29:00 -08:00
e247f141c8
CNTK v2 library: Fix UserDefinedFunction C++ unit tests
2017-01-23 14:59:14 -08:00
4080c0a5af
CNTK v2 library: Enable output shape inference for user-defined functions
2017-01-23 10:53:32 -08:00
373a436cb1
Tools/docker/CNTK-GPU*/Dockerfile: fix base image
...
Also in Tests/Install/linux/Dockerfile-*-GPU.
2017-01-23 15:16:06 +01:00
29bcb50ba8
merged from master (~beta9)
2017-01-22 18:50:15 -08:00
35356ad47e
Tests/EndToEndTests/CNTKv2Python/Examples/CNTK_103B_MNIST_FeedForwardNetwork_test.py: increase tolerance
2017-01-20 17:25:42 +01:00
3ac55535fc
Integrate alrezni/v2_default_unit_gain into master
2017-01-20 02:48:55 -08:00
9b6c6bde34
Add a default (true) for the unit-gain flag value
2017-01-20 09:27:36 +01:00
72292d2207
expose more properties to C#, adapt exampels to use cntkbitmapextensions
2017-01-20 07:19:01 +01:00
ef56278026
Integrate sayanpa/cntk206 into master
2017-01-19 21:12:45 -08:00
91e0a0701c
Integrate kedeng/fixSparseDot into master
2017-01-19 19:14:29 -08:00
1d10e8ab70
Integrate alrezni/mkl_fix into master
2017-01-19 18:19:24 -08:00
6f1909d6b8
Merge branch 'master' of https://github.com/Microsoft/CNTK into sayanpa/cntk206
2017-01-19 13:05:50 -08:00
dcc4b56776
Fix to not output too much in log, which may cause racing condition.
2017-01-19 12:36:12 -08:00
32289b1401
Rename files for Linux test.
2017-01-19 11:59:49 -08:00
76a44b850a
Adding LRN example and associated test. Revised some test to reduce run time.
2017-01-19 11:59:49 -08:00
e6c23fd7da
Fix broken sparse CSC in InnerProductOfMatrices
2017-01-19 11:54:13 -08:00
bcf0e9cb58
Integrate mahilleb/LinPyDefault into master
2017-01-19 11:13:04 -08:00
3c6a697270
Integrate pkranen/deconv2017 into master
2017-01-19 10:39:48 -08:00
f51a85050c
Merge branch 'master' of https://github.com/Microsoft/CNTK into sayanpa/cntk206
2017-01-19 08:42:06 -08:00
82bae8dc09
Force deterministic CPU algorithms for MathTests
2017-01-19 16:40:05 +01:00
7b4f8a9063
Integrate mahilleb/203test into master
2017-01-19 06:53:50 -08:00
7ad880669d
Linux: Py 3.5 default at various places
2017-01-19 15:22:00 +01:00
28efa01cb1
Integrate mahilleb/TestDriverTest2 into master
2017-01-19 05:18:08 -08:00
020869ff17
Integrate alrezni/v2_sweep5 into master
2017-01-19 05:08:00 -08:00
2d91bad7df
CNTK_203_Reinforcement_Learning_Basics_test.py: bump timeout
...
Also, allow a bit more total episodes in fast mode of the notebook.
2017-01-19 13:51:43 +01:00
85c6aa8308
Add end-of-sweep notification
...
- Add sweepEnd flag to the MinibatchData, returned from the
MinibatchSource.
- Add overloads TrainMinibatch, TestMinibatch that take a
Variable-MinibatchData map as their first parameter.
- Add sweepEnd to the Learner::Update function.
- Change the default epochSize to FullDataSweep in
TrainingParameterSchedule ctors.
- Allow minibatches to cross sweep boundaries: add a
flag (a config option) to specify if minibatches are
allowed to cross over the sweep boundaries (which
eliminates partial minibatches at the sweep ends).
- Update python API to pass MinibatchData as is into the
train_minibatch (without replacing it with corresponding ValuePtrs)
- Change the default epoch_size in parameter schedules to None,
in which case the entire data sweep is used as the scheduling unit.
- Add ReaderLib and python unit tests.
2017-01-19 13:11:05 +01:00
cde6154b24
Added test, addressed CR comments
2017-01-19 09:50:23 +01:00
bd9388b700
working version of deconv and unpooling layers in BS
2017-01-19 09:50:22 +01:00
df0ab4e581
Deconv example for master branch
2017-01-19 09:50:21 +01:00
44eb327a62
CNTK v2 library: Fix a bug in Value::Create method for the case when there is a single sequence being packed with a sequenceStart flag setting of false which results in creation of a mask
2017-01-19 00:26:02 -08:00
cf6f50cc26
addressed comments from CR
2017-01-18 19:36:30 -08:00
e8099671a4
Merge branch 'master' of https://github.com/Microsoft/CNTK into sayanpa/cntk206
2017-01-18 17:24:54 -08:00
c6147735a9
Added timeseries with LSTM basics tutorial
2017-01-18 13:12:24 -08:00
119f7de897
CNTK Core: Fix possible Nan accumulation in ElementTimesNode backprop
...
With gradient optimization turned on, node must write gradient without accumulation if ParentOverwritesGradient() is true
2017-01-18 12:33:17 -08:00
6e3811be65
TestDriver.py: Py3 compat changes, add test
2017-01-18 20:44:38 +01:00
a7369b134e
Integrate eldak/trainingSession into master
2017-01-17 10:47:19 -08:00
fc57c33507
Adding python training_session, breaking change inclusing removal of
...
warmup from the minibatch source
2017-01-17 14:38:33 +01:00
dfb5994e5c
Merge remote-tracking branch 'origin/master' into mahilleb/FewerWinDebugTests
2017-01-17 13:35:41 +01:00
c5739da80a
testcases.yml: more consistent formatting
2017-01-17 13:31:36 +01:00
3e834fa94b
Address CR comments
2017-01-17 13:20:12 +01:00
6a2249a99c
Introducing training session object
2017-01-17 10:22:41 +01:00
1690fa89f0
Add samples for Eval V2
...
- add device specifier when loading model
- rename CPPEvalV2Client to CNTKLibraryCPPEvalExamples
- rename CNTKLibraryManagedExamples to CNTKLibraryCSExamples
- rename CNTKLibrarymanagedExamples to CNTKLibraryCSExamples
- add CNTKLibraryCSEvalGPUExamples and CNTLibraryEvalExamples.sln
- complete examples
- update project configuraitons
- rename program.cs to Program.cs; update packages.config
- rename CNTKLibraryCPP/CSExamples to CNTKLibraryCPP/CSEvalexamples
- remove AnyCPU
- add file header, adapt output message
2017-01-16 20:09:59 +01:00
91147fd068
BVT tests: fewer tests on Windows Debug
2017-01-16 17:40:17 +01:00
b6a6ec9203
Integrate eldak/moreDistributionTests into master
2017-01-16 07:17:08 -08:00
7278ad1f41
Adding more tests and fixing bugs in the randomizer/packer
2017-01-16 11:40:12 +01:00
3ab246855b
Add unit-gain momentum flag
...
This commit contains a breaking change: it adds a flag to both
C++ and python Learner API that indicates if the momentum should be applied in
the regular fashion or as a unit-gain filter. This flag is a required
parameter for learners that use momentum (momentum_sgd, nesterov_sgd and
adam_sgd).
2017-01-16 09:39:39 +01:00
43b30a1c41
Integrate chazhang/reader_fix into master
2017-01-13 17:58:30 -08:00
1b0db71ad7
Integrate sayanpa/pycntk105 into master
2017-01-13 16:50:04 -08:00
80321d9148
Added CNTK105 autoencoder tutorial
2017-01-13 11:12:47 -08:00
4c03e2c0a5
Revision based on CR.
2017-01-13 08:52:37 -08:00
07eb5cd576
Modify test result. Verified that the difference in results are due to random seed issue.
2017-01-13 08:52:37 -08:00
85429c1e92
Fix on config files in test.
2017-01-13 08:52:37 -08:00
b28f579131
Bug fix. More config change.
2017-01-13 08:52:36 -08:00
0cc6b42354
More config files change to use the new crop transform.
2017-01-13 08:52:36 -08:00
abc1fc92a7
Bug fixes on crop transform. Change CNTK config files to new crop transform.
2017-01-13 08:52:36 -08:00
4505c48fd0
[CNTK core] sparse GPU impl of FSAdagrad and RmsProp
...
Also fix a bug in dense GPU MultiplyAndWeightedAdd that garbage data may get in via resize
2017-01-12 21:26:56 -08:00
7bb733f445
merged from master (not yet fully building after VS 2015 transition)
2017-01-12 18:47:55 -08:00
9deeb31a1e
Update unittest baseline
2017-01-12 17:43:07 -08:00
f992955cb8
Update the comment.
2017-01-12 17:43:07 -08:00
f9b9070b02
Rename EditDistanceNode to EditDistanceErrorNode. Use GetAllSequences in ED computation
2017-01-12 17:43:07 -08:00
7c8b4f3cb8
Enable Edit distance node in NDL
2017-01-12 17:43:07 -08:00
b4962b728e
Edit distance node
2017-01-12 17:43:07 -08:00
091d3e4421
Integrate mahilleb/DropsPy into master
2017-01-12 12:40:42 -08:00
88b6c328a8
Address CR comments
2017-01-12 15:56:52 +01:00
b3e7bb2e8b
Python: include all wheels in drop
2017-01-12 15:11:19 +01:00
cb66befcf2
CNTK_203_Reinforcement_Learning_Basics.ipynb: add test
2017-01-12 13:59:56 +01:00
4665583810
Make Jupyter Notebook tests device-aware
2017-01-12 13:59:56 +01:00
73f8319b01
Tests/EndToEndTests/CNTKv2Python/Examples/feature_extraction_test.py: fix env variable name
2017-01-12 13:47:09 +01:00
c2d3ee6567
Integrate amitaga/dataInteropFeatures into master
2017-01-11 21:31:20 -08:00
9f894adcf1
Integrate kedeng/BNCuDnn2 into master
2017-01-11 13:46:16 -08:00
546e1d29dc
CNTK v2 library: UserDefined function simplifications a) Add support for saving forward compute values in base BackPropState type itself b) Function::Forward override now receives values for all inputs including Parameters and Constants
2017-01-11 12:01:59 -08:00
fc7331ae6f
added test for feature extraction example
2017-01-11 12:48:43 +01:00
6c73818bf7
[CNTK Core] Change BatchNormalization default to CuDnn in both BrainScript and V2 C++/Python, and add fallback to cntk engine when not supported by cudnn
...
Fix epsilon in CuDnnBatchNormalization to be consistent in forward/backward
Fix BatchNormalization parameter gradient accumulation
Add message on CuDnnBatchNormalization forward infer batch size limitation, and adjust CNTK_202_Language_Understanding eval minibatch size to avoid CUDA failure
2017-01-11 01:00:09 -08:00
7cb1db6620
Merge remote-tracking branch 'origin/master' into mahilleb/vs2015fi
...
Conflicts:
Source/CNTKv2LibraryDll/Utils.cpp
2017-01-10 11:29:03 +01:00
911b2ca82e
Windows: switch to Visual Studio 2015
2017-01-10 10:47:35 +01:00
4b31eda24f
Merge remote-tracking branch 'origin/master' into mahilleb/Mnist13
2017-01-10 08:05:44 +01:00
6dc09565bb
CNTK_103B_MNIST_FeedForwardNetwork.ipynb: add test
...
Also, let run MNIST tests against the default (end-user) data location.
2017-01-10 05:42:30 +01:00
7d236a2d70
Integrate zhouwang/fix-valuebuffer-size into master
2017-01-09 04:40:20 -08:00
1b122c484e
rename CNTKLibraryManagedClient to CNTKLibraryManagedExamples
2017-01-07 10:29:57 +01:00
f7a2172828
move CSEvalV2Examples to the examples folder and move the project files to the Test folder
2017-01-07 10:29:57 +01:00
750dccbfd3
remove isResizable and SequenceLength parameters from CopyTO until we have clear understanding for the use case; rename CopyTo to CopyVairiableValueTo; adapt test cases
...
fix error
fix errors in tests; add tests for exceptions
add more check in GetSequenceAndBatchLength(); Address CR comments
2017-01-06 10:06:49 +01:00
69c75c4494
Using Variable as parameter to CopyTo to infer dynamic axes correctly; Add check for PackedValue; Simplify processing based on assumption that sequence starts at 0 and ends at the first invalid mask; other changes from CR.
...
add tests for variable with 0,1 or 2 dynamic axes, and adapt tests to new API, refactor tests to share code
restore main and tests
make class Value after class Variable to make linux happy
fix bug in test
2017-01-06 10:06:48 +01:00
88ecb8d4ff
update CopyTo for OneHot using vocabSize instead of shape
2017-01-06 08:53:27 +01:00
6ff0427195
add unit tests for CopyTo, improve unit tests to use random sequences, add test for single sample, single sequence in Value.Create
...
fix linux build
add missing error message.
fix linux build
switch order of function definition to make linux happy
add instantiation of functions
minor changes
add comments, using random number in tests
restore main.cpp
2017-01-06 08:53:27 +01:00
f1cbc08e94
add CopyTo for onehot vector; refactor code to share between CopyTo methods;
...
fix bug using GPU device. The cpuArrayView must be alive in the function scope
add CopyTo for OneHot; support Value in both dense and sparse format
2017-01-06 08:53:27 +01:00
f5fdbeb028
refactor Value::CopyTo to allocate memory in .h file; adapt unit tests; add sequencesLens as parameter for Value::CopyTo
2017-01-06 08:53:26 +01:00
6290e8e9a7
Integrate nikosk/assignsumof into master
2017-01-05 20:10:09 -08:00
a026a272bf
addressed cr feedback
2017-01-05 18:24:32 -08:00
c5527720b4
merged from amitaga/blockFunctionFixes
2017-01-05 11:35:22 -08:00
8aa380bd24
Allowing deserializers return invalid sequences
2017-01-05 11:05:58 +01:00
b7c031cc16
Merge remote-tracking branch 'origin/master' into nikosk/assignsumof
2017-01-04 16:40:39 -08:00
59dcc28af5
Some more self-assignment unit tests just in case
2017-01-04 16:39:50 -08:00
78c051ec90
CNTK core: Turn on forward compute memory sharing by default
2017-01-04 15:20:16 -08:00
7308637280
fix bug in ValueBuffer constructor: setting Size correctly; Add tests
...
update baseline file
2017-01-04 16:30:00 +01:00
2b5622ca36
merged from wdarling/lstmaux to get s2s sample (nor working due to has_aux change though)
2017-01-03 10:31:42 -08:00
bec6a499b0
Integrate eldak/sequenceLocalTimelineBySample into master
2017-01-03 07:23:36 -08:00
f39317a5aa
Adding local timeline
2017-01-03 14:22:36 +01:00
8801a604b6
update the test to use model version 18
2017-01-03 10:41:00 +01:00
Frank Seide
Nathan Luehr
Thilo Will
Vadim Mazalov
Mark Hillebrand
Zhou Wang
Guenther Schmuelling
Philipp Kranen
Eldar Akchurin
REDMOND\sayanpa
Amit Agarwal
Project Philly
Ivan Rodriguez
PengchengHe
Cha Zhang
Alexey Reznichenko
missxa
KeDengMS
Nikos Karampatziakis