12a0220b15
CNTK v2 library: Cleaned up python examples and added some inline documentation
2016-09-05 06:04:56 -07:00
3f9f248146
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/cntkv2LibraryOperators
2016-09-03 17:31:11 -07:00
abf468097b
CNTK v2 library: Trainer checkpointing and some MinibatchSource creation helper APIs
2016-09-03 17:31:02 -07:00
dae3967e05
CNTK v2 library: Some API simplifications and initial stab at trainer checkpointing (not yet functional)
2016-09-02 06:12:07 -07:00
9afffea658
rebase on master and adjust
2016-09-01 14:07:03 +02:00
7a5c133edc
CNTK v2 library: Added Seq2Seq implementation as a test using V2 C++ API and other related changes
2016-08-30 22:50:25 -07:00
4299d870c8
CNTK v2 library: API simplifcation, added new primitive Functions and addressed pending CR feedback
2016-08-26 18:29:30 -07:00
66498cf414
Merge remote-tracking branch 'origin/master' into mahilleb/CuDnn5Test
...
Note: baselines need to be fixed for
Tests/EndToEndTests/BatchNormalization and
Tests/EndToEndTests/Examples/Image/Miscellaneous/CIFAR-10/02_BatchNormConv.
2016-08-23 11:12:35 +02:00
f76afa2b7e
Switch to CuDNN v5
...
For batch normalization, running inverse standard deviation becomes
running variance. We mirror this CuDNN v5 change in the CNTK batch
normalization engine. Model version is bumped. When old models are
loaded, this parameter is (approximately) converted.
In the same model version change, let batch normalization count
samples seen rather minibatches (this deals with incorrect averaging
when minibatch size is varied across epochs).
For batch normalization averaging and blending handle initialization
cases, don't rely on mean and variance initial values (set in
NDL/BrainScript).
Update Windows / Linux / Docker build.
With this commit, CuDNN v4 is not supported anymore.
2016-08-22 17:55:10 +02:00
668b6a467f
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/cntkv2Library
2016-08-21 03:52:37 -07:00
fa4b99d102
CNTK v2 Library: a) Add dynamic axis support b) New primitive functions and some higher level functions and c) Sequence classification test
2016-08-21 03:49:03 -07:00
9bd9308d2e
Improve V2 Learner API, add basic serialization.
...
FSAdaGrad, RMSProp: pre-allocate smoothed gradients with expected
number of columns.
Improve configuring built-in learners: add learning rates and
momentums schedules.
Add NDArrayView as a DictionaryValue type.
Add tests for serialization and basic learner functionality.
2016-08-18 15:06:54 +02:00
a5614771ec
adapt swig wraper to the new getNextMiniBatch
2016-08-09 16:14:19 +02:00
2e6a7116b2
merge with master
2016-08-08 15:12:30 +02:00
573fab1d25
first training run with readers config
2016-08-02 16:52:12 +02:00
0854708400
CNTK v2 library: a) Convolution b) Mean variance normalization c) Added Cifar resnet test
2016-07-30 16:16:41 -07:00
4a8be29046
temporarly comment one of Variable constructors, and add initial wrapper for vector<FunctionPtr>
2016-07-27 15:21:15 +02:00
b119474d49
resolve conflicts after merging with master
2016-07-27 10:12:08 +02:00
fcc9d09122
Cherry-pick from amitaga/cntkv2Library: CNTK v2 library: Add member method to NDMask to enable retreiving the buffer underlying a Mask object
...
Conflicts:
Source/CNTKv2LibraryDll/NDMask.cpp
2016-07-26 14:23:31 +02:00
759fa18a7c
CNTK v2 library: Add member method to NDMask to enable retreiving the buffer underlying a Mask object
2016-07-25 22:43:11 -07:00
3e6926deb0
Merge branch 'master' of https://github.com/Microsoft/CNTK into amitaga/cntkv2LibraryOperators
2016-07-25 16:52:15 -07:00
a632393c11
CNTK v2 library: Expose 18 additional operators in the v2 API
2016-07-22 18:41:48 -07:00
5b1c217688
CNTK v2 library: Added initial Trainer and MinibatchSource implementations, enabling end-to-end model training with v2 API. A test added to train a feed-forward network on the MNIST dataset using the v2 C++ API
2016-07-22 16:46:09 -07:00
48379019bf
fixed needed to run swig after merge with master
2016-07-21 17:45:04 +02:00
60bcf71314
merge with master, remove tabs
2016-07-21 14:32:42 +02:00
a79f6b3c9a
merge with master
2016-07-21 14:17:19 +02:00
9ba93ab84e
CNTK v2 library: Added support for loading legacy v1 format models and saving models in v1 format
2016-07-17 16:44:44 -07:00
d6c9239478
Remove unused comparison operator for Variable
2016-07-15 15:14:03 +02:00
1e303cc345
Supporting unordered_* types; plus tests pass (minus one broadcasting test)
2016-07-14 19:17:16 +02:00
1b0548fdde
Add v2 Learners (standalone)
2016-07-13 11:10:02 +02:00
f4031557fd
Adapting to new STL based API
2016-07-13 00:14:20 +02:00
72fefc1030
Putting back explicit,final, and static_assert guarded by SWIG preprocessor macro
2016-07-12 20:03:45 +02:00
f9253083d8
Prepare API for backward adjustments
2016-07-12 20:01:56 +02:00
e2d799d7e9
Cleaned up after merge
2016-07-12 20:00:17 +02:00
39cfd404e9
First working Swig bindings for Python
2016-07-12 19:59:14 +02:00
ce163fd23d
CNTK v2 library: Fix linux build
2016-07-08 18:14:05 -07:00
415706a678
CNTK v2 library: a) Removed STL type wrappers and directly use STL types in API b) Address the issue of inlined functions getting exported by moving the export specification to a per function level
2016-07-08 17:56:44 -07:00
828668265d
CNTK v2 library: Address CR feedback
2016-07-07 21:18:20 -07:00
b26b8bc24e
CNTK v2 library: Addressed CR feedback
2016-07-06 01:30:07 -07:00
382535baf8
CNTK v2 Library: Address CR feedback
2016-06-15 11:58:32 -07:00
26c2006541
CNTKv2Library: Linux port
2016-06-11 16:50:26 -07:00
c508e9f4ca
CNTK v2 library: Sparse input support
2016-06-11 02:12:39 -07:00
933adb275a
Add recurrence and variable length sequence support for the 'Function' abstraction
2016-06-05 14:36:16 -07:00
76bf193267
First cut of the CNTK v2 C++ library implementation
2016-05-31 06:40:40 -07:00
Amit Agarwal
jeanfad
Mark Hillebrand
Alexey Reznichenko
Willi Richert