LAN-PyTorch

Implementation of Logic Attention Network in PyTorch (currently without inductive test)

Notes:

1. This repo currently does not support inductive testing, e.g. aux and dev data in the original repo are not needed and removed.
2. This repo currently contains only Attention aggregator and TransE score.
3. L2 regularization is temporarily removed.

The model must be trained with GPU.

Data requirement

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!)

Evaluation Methods

Filter out the known entities (i.e., the ones in train set)