An easy-to-use sequence labeling project(get SoA on ATIS data) with pytorch
Python
Pytorch: Bi-LSTM Slot-Fillig
Introduction
A simple sequence labeling with bi-lstm.
ATIS data is used and included.
Requirement
- Python 3.5+
- Pytorch 0.3+
- pytorchnlp
Tricks Included
- glove embedding
- mini-batch
- learning rate decay
- grad clip
- dropout
- sgd & adam
Usage
Train and run the model in the project dir.
python3 postagger.py --train_an modest'md_test --train_path ./data/atis_train --dev_path ./data/atis_dev --test_path ./data/atis.test.txt --label_set_path ./data/atis_slot_names.txt --model ./model/ --output ./result/result.txt --script ./eval/conlleval.pl --gpu 0