Boxes_for_Joint_hierarchy_AKBC_2020

Setup

1. Clone the repo

2. Install the requirements:

pip install -r requirements.txt

Getting the preprocessed data

The data used in the paper can be found here.

Reproducing the results for box embeddings

Without Weights&Biases logging

export DATA_DIR=directory/where/unzipped/data/folder/is
export CUDA_DEVICE=0  # =-1 for cpu
export WANDB=false 
allennlp train model_configs/hypernym_0.jsonnet --serialization-dir hypernym_0_training_dump --include-package=datasets --include-package=boxes --include-package=models

With Weights&Biases logging

Assuming your username and project is username and project respectively.

export DATA_DIR=directory/where/unzipped/data/folder/is
export CUDA_DEVICE=0  # =-1 for cpu
export WANDB=true 
wandb_allennlp --subcommand=train --config_file=model_configs/hypernym_0.jsonnet --include-package=datasets --include-package=boxes --include-package=models --wandb_entity=username --wandb_project=project --wandb_run_name=hypernym_0

Using the trained models on test data

export DATA_DIR=directory/where/unzipped/data/folder/is
python predict_f1_test.py --model hypernym_0_training_dump 

Replace hypernym_0 with hypernym_{10,25,50}, meronym_{0,10,25,30} and joint to train all the regularized box models reported in the paper.

If you use the code, please site the following paper.

@inproceedings{
patel2020representing,
title={Representing Joint Hierarchies with Box Embeddings},
author={Dhruvesh Patel and Shib Sankar Dasgupta and Michael Boratko and Xiang Li and Luke Vilnis and Andrew McCallum},
booktitle={Automated Knowledge Base Construction},
year={2020},
url={https://openreview.net/forum?id=J246NSqR_l}
}