Skip to content

BotGNN: Inclusion of Domain-Knowledge into GNNs using Mode-Directed Inverse Entailment

License

Notifications You must be signed in to change notification settings

tirtharajdash/BotGNN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

22 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Inclusion of Domain-Knowledge into GNNs using Mode-Directed Inverse Entailment

Graph Neural Network directly constructed from the Bottom-Clause generated by Inductive Logic Programming (ILP). The resulting GNN models are to be known as "BotGNNs" (singular, "BotGNN").

Data

The present uncompressed version of our processed data constitutes almost 40GB. We have now released all the GNN-ready datasets (n=73) via Open Science Framework (OSF). Here is the link: BotGNN Data.

Old: We will release these data as after the acceptance of our paper. However, in the mean time, we will release a sample dataset as an example.

Experiments

The MDIE implementation is carried out within Aleph System. The GNN implementations closely follow the implementations in VEGNN. Also, VEGNN forms a yard-stick for comparison of BotGNNs.

The directory structure is a bit different here. We conduct numerous experiments in this work, over 80 different datasets and with 5 different variants of GNNs. Therefore, for convenience, we keep each experiment separately under each dataset directory. Please refer to the directory structure here.

(More details on this repo will be added soon. Keep watching!!!)

Updates

[19 May 2023] All the 73 datasets are shared via Open Science Framework (OSF).
[27 May 2021] Repository is made public.
[25 May 2021] The preprint of the article is available online via arXiv.
[25 May 2021] The article is submitted to MLJ.
[28 Sep 2021] The article got accepted at MLJ.
[18 Nov 2021] The article is now published.

How to cite our work and data

@article{dash2021inclusion,
	author="Dash, Tirtharaj
	and Srinivasan, Ashwin
	and Baskar, A.",
	title="Inclusion of domain-knowledge into GNNs using mode-directed inverse entailment",
	journal="Machine Learning",
	year="2022",
	month="Feb",
	day="01",
	volume="111",
	number="2",
	pages="575--623",
	issn="1573-0565",
	doi="10.1007/s10994-021-06090-8",
	url="https://doi.org/10.1007/s10994-021-06090-8"
}

To cite data, if you use them:

@article{Dash2023BotGNNdata,
	title={NCI Datasets (n=73) for BotGNNs},
	html={https://doi.org/10.17605/OSF.IO/HNZ5R},
	DOI={10.17605/OSF.IO/HNZ5R},
	journal={Open Science Framework},
	publisher={OSF},
	author={Dash, Tirtharaj},
	year={2023},
	month={May},
	abbr={OSF}
}