bilstm-crf sequence labeling model

Quick Start: Demo

install deeptext and train your sequence labeling model to start sequence labeling!

before running the command below, make sure your data for model training, validating or testing are in right format.

Run command below to train a model:

python run.py -m train -o your/model_dir -d you/train_data_file -v your/valid_data_file`

Run command below to test your model:

python run.py -m test -o your/model_dir -d you/test_data_file`

Run command below to drop into an interactive session. For each sentence, model will return the sequence labeling result string.

python run.py -m try -o your/model_dir`

Install Deeptext

Setting up Deeptext is easy!

Deeptext requires Linux/OSX and Python 2.7. Its other dependencies are listed in requirements.txt. CUDA is strongly recommended for speed, but not necessary.

Run the following commands to clone the repository and install Deeptext:

git clone https://github.com/liuxiaoan8008/deeptext.git
cd deeptext; pip install -r requirements.txt

Data Format

train, valid or test data file format:

e.g. train_data.txt

file content:

^ 第 一 天 ： 上 午 桂 林 出 发 前 往 资 源 县 城 $
O O O O O O O E E O O O O O O O O O
^ 全 县 平 均 海 拔 在 8 0 0 米 以 上 $
O O O O O O O O O O O O O O O
^ 是 典 型 的 高 寒 山 区 $
O O O O O O O O O O
^ 全 县 气 候 温 和 $
O O O O O O O O

Training Options

Optional arguments:

  -s    STEP,      train step, default is 500
  -b    BATCH,     train data's batch size, default is 256
  -l     LEN,      max document len, default is 25
  -e    EMBED,     word2vec embedding size, default is 100
  -p    PROB,      dropout prob, default is 0.5