Implementation of Logic Attention Network in PyTorch (currently without inductive test)
Notes:
- This repo currently contains only Attention aggregator and TransE score.
- L2 regularization is temporarily removed.
The model must be trained with GPU.
Dataset should be put under data directory, and it should contain files including train, test, entity2id.txt and
relation2id.txt. To translate the data in index format, see utils.preprocess (the function will overwritten the original files!)
Filter out the known entities (i.e., the ones in train set)