cxjtju / ta_transe

Temporal Knowledge Graph Completion | Implementation of Temporal Knowledge Graph Completion following the work of Duran et al., Learning Sequence Encoders for Temporal Knowledge Graph Completion

Code framework for Temporal KG completion.

Implementation of Temporal Knowledge Graph Completion following the work of Duran et al., Learning Sequence Encoders for Temporal Knowledge Graph Completion

• Code:

  • modify based on knowledge_representation_pytorch

  • 1st column in train.txt - subject entity

  • 2nd column - relation

  • 3rd column - object entity

  • 4th column - time

  • 1st figure in stat.txt - number of entities

  • 2nd figure in stat.txt - number of relations

  used preprocess_TA_step1.py and preprocess_TA_step2.py to make data for TA_TransE.

  python preprocess_TA_step1.py ICEWS14
  python preprocess_TA_step2.py ICEWS14
  

  Note : data is already Preprocessed. if you have new data then you can follow the above Process i.e. "python preprocess_TA_step1.py ICEWS14"

• TATransE.py : train code

• You can run the code with

   python TA_TransE.py ICEWS14
   
  

  eg:

   cd ./baselines
   CUDA_VISIBLE_DEVICES=0 python TA_TransE.py -f 1 -d ICEWS14 -L 1 -bs 1024 -n 1000
  
  

Note: when you run the code like python "TA_TransE.py ICEWS14", you need to give dataset in argument with out _TA