* More extensive examples of graph building.

* Review comments

* Review comments

* Review comments

* WIP support bn training for global_stats

(cherry picked from commit eb81a37328ea177b1d58c9eebdbb345e0fa25f0d)

* - Style fix
- Fix test case

* Addressed PR comments
- added support for bn training/inference with a same ctor
- added more verbose comments in bn header

* Fixed bn serializer and default value in bn ctor for bwd compatibility

* proposed docs change

* - Addressed PR comments
  - added support to compute bn inference/training using same mkldnn kernel with global stats

* fix unit bn relu unit test

* add random init for all input types, not just float

* remove debug

* initial refactoring using PM

* unit test pass

* cosmetic changes

* add another rnn test

* address louis' feedback

* lower-case labels

* Add python wrappers for nGraph Cos, Cosh operations.

- Update docstrings.

* Enable auxiliary function running computation to accept multiple input nodes.

* Python wrapper for nGraph Dot function with UT.

* Update python wrappers for nGraph Exp and Equal operations.

- Update docstrings.
- Add UT for exp.

* Update python wrappers for nGraph Floor, Greater, GreaterEq, Less, LessEq operations.

- Update docstrings.
- Add UT for ng.floor.

* Update python wrapper for nGraph Log operation.

- Update docstring.
- Add UT.

* Change serialize routine

* Change serialize routine call for onnx

* Update function type annotation.

* Update avg_pool signature to be consistent with ngraph AvgPool.

- Change parameters order and use more appropriate paramter names.
- Add docstring.

* Single file with test for pooling operations.

- Add test for avg_pool for 2D case.

* Code refactoring.

- Rename parameters to be more verbose.
- Change function return type, to enable general usage.

* Add UT for ceil, ceiling and abs.

* Update docstrings and type annotations.

* Add UT for broadcast operation.

* Add UT for concat operation.

- Minor change: add optional node name function parameter.

* Code formatting.

* UT for constant and convert operations.

- Move broadcast test to test_basic.py file.

* Update function signature.

- Update to be consistent with nGraph object API.

* Review fix.

- Update type annotations.
- Update docstring.
- Change local variables names to be consistent.

* Refactoring - put some commonly used functions into util.py.

* Update convolution operation signature.

- Add docstring
- Update to be consistent with respective nGraph object API.
- Formatting in UT.

* Fix import statements under py27.

* Correct quotation marks.

* Revert changes: use AxisSet as a return type.

* Review fix.

- Update parameter names to be consistent with respective parameters of nGraph object constructors.

* Review fix

- Set seed for random number generation to be repeatable.
- Use numpy.allclose.

* Change serialize routine

* Change serialize routine call for onnx

* Run clang-format on constant.cpp

* Update function type annotation.

* CPU: Rename PREFER_EIGEN to something clearer

* More renaming

* move to test_nnp_backend.cpp

* format

* link libnnp.so in unittest

* making replace_node more robust

* fix error msg

* tweak err msg

Fix LLVM commit hashes to work around recent version bump incompatible with our hard-coded "5.0.1" paths (#781)

Changes to layout selection heuristic for cases like XLA frontend where fprop layouts are not propagated to bprop graph (#769)

* fix relu(conv); the old test was fusing conv(relu) instead of running both separetely

* fix segfault due to initiating a vector of result vectors w/ a number of arguments

* [Py] Update serialization

* Update docstring

* Add UT for serialize model

* fixed two bugs.

* refactor infer shape.

* clang format.

* reverted some changes.

* moved infer_convolution_output_shape() to outside class

* moved infer_convolution_output_shape to op::util namespace

*  split cpu_fusion into cpu_fusion and cpu_test; clean up headers

* fix formatting

* add new line to tensor_mask.hpp

* move CoreFusion to ngraph/pass

* fix incorrect header path

* add nd reshape

* compiler and no crash with wrong result version

* change output_stride to trans_stride, which transform input idx to output idx

* using vector instead of c array

* remove delete

* using const and reference to pass string and array

* change 'unimplement' comments, remove extra indents

* format and cast size_t to int