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.

1. get eigen3
2. open git bash or cmd, change directory to the repository root
3. git submodule init && git submodule update
4. make a directory to build, mkdir build
5. cd build
6. 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.
7. 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 :

1. postagging based on example tag-bilstm.cc of DyNet [done]

2. chinese segmentation(using sequence labeling method) , ner [done]

3. more various structures based on DyNet [doing]

4. (almost)from scratch ?? -> NO , need more time to think about it !

WIKI

wiki pages for more detail infomation.