Relation Classification

Installation

1. My Code:

• Frist open google colab with GPU

• Clone this project to Google Colab:

  !git clone https://github.com/dqhuy140598/RelationClassification.git

• Download pretrained Word2Vec:

  !wget https://s3.amazonaws.com/dl4j-distribution/GoogleNews-vectors-negative300.bin.gz

• Unzip pretrained Word2Vec:

  !gunzip GoogleNews-vectors-negative300.bin.gz

• Cd to this project:

  %cd RelationClassification/

• Generate training and validation shortest dependency path and part of speech tagging:

  !python ./utils/data_utils.py

  (data/processed/)

• Build words vocabulary:

  !python ./utils/build_vocab.py

  (data/processed/vocab.txt)

• Train model:

  • If you use threshold equal to 0.5:

    !python train.py --pretrained $PRETRAINED_PATH$

  • Else:

    !python train.py --pretrained $PRETRAINED_PATH$ --use_thresh True

  • $PRETRAINED_PATH$ : PATH to pretrained Word2Vec. Please replace it by /content/GoogleNews-vectors-negative300.bin

2. EMNLP2017 Code:

• Convert the processed data to use for this code:

  !python ./utils/convert.py

• Train model:

  • If you use threshold equal to 0.5:

    !python EMNLP2017.py --data $DATA_PATH$ --pretrained $PRETRAINED_PATH$

  • Else:

    !python EMNLP2017.py --data $DATA_PATH$ --pretrained $PRETRAINED_PATH$ --use_thresh True

  • $PRETRAINED_PATH$ : PATH to the pretrained Word2Vec. Please replace it by /content/GoogleNews-vectors-negative300.bin

  • $DATA_PATH$: PATH to the converted data. please replace it by data_convert.json