Skip to content

Latest commit

 

History

History
39 lines (31 loc) · 1.88 KB

README.md

File metadata and controls

39 lines (31 loc) · 1.88 KB

VOE: Vector-Based Order Embedding

Our pytorch implementation of the research work "Order-Embeddings of Images and Language". Order embedding or order representation learning (VOE) is a must for applicaitons like pre-requiste relationship learning. Please refer to the original codebase in Lua for more details.

Dependencies

The core learning model is built using PyTorch

  • Python 3.6.3
  • PyTorch 0.3.0

To learn order embeddings

  python main.py -d <name> -h1 <dimension> -bs <batch_size> -e <#epoch> -gpu <device>

Data

There should be five data files ready in the 'datasets' folder, e.g. datasets/name/

  • <name>_split_train.pkl list of training instance in pickle format, each instance is a three tuple: (ent1, ent2, label), label is either -1 or 1
  • <name>_split_dev.pkl list of validation instance in pickle format, instance (ent1, ent2, label) is a three tuple
  • <name>_split_test.pkl list of testing instance in pickle format, instance (ent1, ent2, label) is a three tuple
  • <name>_uid_userDoc.npy one hot encoding of the entities in numpy format
  • <name>_iid_itemDoc.npy one hot encoding of the entities in numpy format

Cite

Please consider cite our paper if you find the paper and the code useful.

@inproceedings{chiang2019one,
  title={One-class order embedding for dependency relation prediction},
  author={Chiang, Meng-Fen and Lim, Ee-Peng and Lee, Wang-Chien and Ashok, Xavier Jayaraj Siddarth and Prasetyo, Philips Kokoh},
  booktitle={Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval},
  pages={205--214},
  year={2019}
}

Feel free to send email to [email protected] if you have any questions. This code is modified from ANR.