README.md

KBI-Regex

This repository contains code and data for reproducing the results of our paper:

Vaibhav Adlakha, Parth Shah, Srikanta Bedathur, Mausam. Regex Queries over Incomplete Knowledge Bases

To cite this work, please use the following citation:

@inproceedings{
adlakha2021regex,
title={Regex Queries over Incomplete Knowledge Bases},
author={Vaibhav Adlakha and Parth Shah and Srikanta J. Bedathur and Mausam .},
booktitle={3rd Conference on Automated Knowledge Base Construction},
year={2021},
url={https://openreview.net/forum?id=4YQVfA5vEJS}
}

Downloading Datasets

The paper introduces two datasets for regex queries over incomplete knowledge bases - FB15K-Regex and Wiki100-Regex. These datasets are available to be downloaded from the following links -

1. FB15K-Regex
2. Wiki100-Regex

Each dataset is a .zip file. This needs to be placed in a cache/raw-datasets directory within the project directory. The directory structure should look like the following:

kbi-regex
     |___ cache
            |___ raw-datasets
                     |____ fb15k_regex_data.zip
                     |____ wiki_v2_regex_data,zip

The zip files will be unpacked and encoded when the code is run for the first time, and will be used for subsequent runs.

Running Models

To run the model, we first need to install the repository as a package. From the project directory, run the following command:

pip install --no-index --no-deps -e .

Use the following command to run RotatE-Box (Free parameter + Aggregation) on FB15K-Regex

python main.py \
    --model RotatE --box \
    --kleene_plus_op free_param \
    --dataset fb15k \
    -dim 800 \
    --query_types "0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 21" \
    --do_train \
    --learning_rate 1e-4 \
    --lr_schedule half \
    --truncate_loss \
    --batch_size 1024 \
    --negative_sample_count 256 \
    --max_epochs 500 \
    --resume_from_checkpoint <PATH TO KBC MODEL> \
    --seed 0 \
    --num_workers 8 \
    --save_dir <PATH TO OUTPUT DIR>

Contact

For queries and clarifications please contact vaibhav.adlakha (at) mila (dot) quebec