* "make global tensor function independently"

* "replace functor"

* "fix inline template error"

* "fix tensor array with CopyFrom"

* "fix other case use CopyFrom"

* "move the op interface hardly"

* "fix operators"

* "fix typo"

* "delete dynamic recurrent rnn and fix gru_unit in debugmode"

* "fix unique_ptr copy"

* "fix cuda copy"

* "fix namespace error"

* "removed nccl python test"

* "fix include error"

* "fix typo"

* fix copy util test

py unit test should use correct executable file

Roi pool operator

* beam_search_decode support multi data type

* add VisitDataType for beam search decode

* use Specialization to handle bool

* move Specialization of BeamSearchDecodeFunctor out of class

* is_training to is_test in dropout op

* handle dropout and batch_norm operator when prune pdesc in testing mode

* handle dropout and batch_norm operator when prune pdesc in testing mode

* add get_inference_program method

* fix dropout op

* fix ci

* test data after each batch training

* refine code

* refine test_book3

* fix ci

* follow comments

* make enforce a target and dependent on nccl when gpu is enabled

* add some more dependency

* Change all `data_type` in Python to `dtype`

* Change `date_type` in C++ to `dtype`

* Refine

Fix latex syntax in  bilinear tensor product operator comment.

fix LaTeX syntax in three operators' comments.

 Fix lstm_op and gru_op in debug mode.

Fix LaTeX equation for huber_loss_op.cc.

* Support int64 for sum op

* Refine code

remove vector::erase

* Adding the FTRL optimizer

* Fixed the python test case

* Updating the writeup of RNN doc

* Editing the documentation for seq_decoder, and fixing typos

* Fixing the captioning on 2 level RNN

* pushing after a pull

* Editing and re-writing parts of Data Reader design doc

update googlenet benchmark data

fix v2 init issue on Mac

Make epsilon in BatchNormLayer a configurable variable.

Bug fix in dense example of CAPI.

* init label_semantic_roles.py

* add linear_chain_crf and test

* complete test_linear_chain_crf

* correct last layer of db_lstm

* update optimizer and initializer

* update param_initializer of embedding_layer

* support load pre trained embedding

* rm unused parameter

* optimize code

* clean code

* fix test

* add todo

Fix calculations in gru_unit_op to consistent with gru_op 

remove the tmp buffer when merge grad

Implementing the MSRA initializer for rectifier units