Sequence Labeling by Neural Network
The repository is for sequence labeling through neural network (deep learning) methods .
RUN FLOW SAMPLE
directory run_flow_example contains naive example to bulid project and run samples under linux .
Build
dependency
we are using DyNet library fork DyNet-self (some trivial modified) as the basic neural framework. After clone the repository, we should use
git submodule init
git submodule update
to clone down the dynet
module.
Dynet needs boost
and eigen3
. cmake
is also needed.
Under MSVC
boost-1.57.0, boost-1.58.0 are supported, and boost-1.60.0 leads to some compiling errors.
- get eigen3
- open
git bash
orcmd
, change directory to the repository root git submodule init && git submodule update
- make a directory to build,
mkdir build
cd build
- using the command to make :
cmake .. -DEIGEN3_INCLUDE_DIR=/eigen/path -DBOOST_ROOT=/boost/path -DBoost_USE_STATIC_LIBS=On
, Boost_USE_STATIC_LIBS=On is needed for Windows. - open the VS solution under build folder
Under Linux
you can just use run_flow_example
Plan
it is now based on DyNet library
steps :
-
postagging based on example
tag-bilstm.cc
of DyNet [done] -
chinese segmentation(using sequence labeling method) , ner [done]
-
more various structures based on DyNet [doing]
-
(almost)from scratch ?? -> NO , need more time to think about it !
WIKI
wiki pages for more detail infomation.