Easy Start

April 22, 2025 · View on GitHub

English | 简体中文

Requirements

python == 3.8

  • torch == 1.5
  • hydra-core == 1.0.6
  • tensorboard == 2.4.1
  • matplotlib == 3.4.1
  • scikit-learn == 0.24.1
  • transformers == 3.4.0
  • jieba == 0.42.1
  • deepke

Download Code

git clone https://github.com/zjunlp/DeepKE.git
cd DeepKE/example/re/standard

Install with Pip

  • Create and enter the python virtual environment.
  • Install dependencies: pip install -r requirements.txt.

Train and Predict

  • Dataset

    • Download the dataset to this directory.

      wget 121.41.117.246:8080/Data/re/standard/data.tar.gz
tar -xzvf data.tar.gz

    • Three types of data formats are supported,including json,xlsx and csv. The dataset is stored in data/origin:

      • train.csv: Training set
      • valid.csv : Validation set
      • test.csv: Test set
      • relation.csv: Relation labels

  • Training

    • Parameters for training are in the conf folder and users can modify them before training.
    • If using LM, modify 'lm_file' to use the local model.
    • Logs for training are in the log folder and the trained model is saved in the checkpoints folder. This task supports multi card training. Modify trian.yaml's parameter use_multi_gpu to true, gpu_ids set to the selected gpus. The first card is the main card for calculation, which requires a little more memory.show_plot set to visualize the loss of the current epoch.The default value is False.
    python run.py

  • Prediction

    Set the fp in predict.yaml as the path of the trained model / checkpoint to be used in prediction.The absolute path of the model needs to be used，such as xxx/checkpoints/2019-12-03_ 17-35-30/cnn_ epoch21.pth.

    python predict.py

Models

  1. CNN
  2. RNN
  3. Capsule
  4. GCN (Based on the paper "Graph Convolution over Pruned Dependency Trees Improves Relation Extraction")
  5. Transformer
  6. Pre-trained Model (BERT)

Data Labeling

If you only have sentence and entity pairs but relation labels, you can get use our distant supervised based relation labeling tools.

Please make sure that:

  • Use the triple file we provide or high-quality customized triple file
  • Enough source data