Graph-aware Logistic Regression

This repository contains the source code to reproduce experiments in the paper: Graph as a Feature: A Graph-Aware Non-Neural Model for Node Classification.

Usage

The src/main.py file can be used with the following parameters: For a specific model, the src/main.py can be used with the following parameters:

--dataset         Graph dataset {cora, pubmed, ...}
--undirected      If true, force graph to be undirected 
--randomstate     Random state seed (default=8)
--k               Value for k-fold splits
--stratified      If true, perform stratified split in training folds (default=true)
--model           Model name {Diffusion, GCN, ...}
--use_features    [optional] If true, use features as input (for baseline models)
--use_concat      [optional] If true, use concatenation of adajcency and feature matrices as input (for baseline models)

For example, running expriment on Cora for the Graph-aware Logistic Regression model is done with:

python main.py --dataset=cora --undirected=true --penalized=true --randomstate=8 --k=8 --stratified=true --model=Logistic_regression --use_concat=true

Datasets

We use the following graph datasets in our experiments. All datasets are available in the \data directory.

Dataset	$$|V|$$	$|E|$	$L$	$C$	$\delta_A$
Cora	2708	10556	1433	7	2.88e-03
Pubmed*	19717	88651	500	3	4.56e-04
Citeseer	3327	9104	3703	6	1.65e-03
Actor	7600	30019	932	5	1.04e-03
CS	18333	163788	6805	15	9.75e-04
Photo	7650	238162	745	8	8.14e-03
Cornell	183	298	1703	5	1.79e-02
Wisconsin	251	515	1703	5	1.64e-02
Wikivitals	10011	824999	37845	11	8.23e-03
Wikivitals-fr	9945	558427	28198	11	5.65e-03
Wikischools	4403	112834	20527	16	5.82e-03
Wikivitals+	45149	3946850	85512	11	1.93e-03
ogbn-arxiv	169343	1166246	85512	40	8.14e-05