This project applied bi-lstm+crf model into named entity recognition task, which belongs to sequence labeling.

• Python3.5+
• PyTorch 0.4+
• numpy
• tensorboardX

• utilize torch.utils.data.Dataset, torch.utils.data.Dataloader to load corpus

• conditional random field module inside

• evaluate f1 on entity

• support both cpu & gpu

• user-friendly tensorboard visualization

• abundant arguments for tuning parameters

• nice code style & elaborate comments

for each line

token1 token2 token3 token4, tag1 tag2 tag3 tag4

please refer to test.csv for more detail.

export CUDA_VISIBLE_DEVICES=X
python main.py --model-name MODEL_NAME --vocab token2idx.json tag2idx.json

python main.py --model-name MODEL_NAME --vocab token2idx.json tag2idx.json --no-cuda

• support customized dataset
• support multi-gpu training
• refine viterbi decoding efficiency
• extract evaluate mode