/TDAR

This is a Pytorch implementation of our "Topology-Driven Attribute Recovery".

Primary LanguagePythonMIT LicenseMIT

TDAR

This repo is for source code of paper "TDAR: Topology-Driven Attribute Recovery for Attribute-missing Graph Learning". This code is referenced from SVGA and SAT, thanks to the author's contribution.

Environment Settings

python==3.11.4
scipy==1.10.1
torch==2.0.1
torch-geometric==2.3.1
numpy==1.24.3
scikit_learn==1.3.0
GPU: GeForce RTX 4090

Dataset

The data obtained from link is approximate.

Usage

You can use the following command to run our model.

python main.py --data cora --lr 1e-3 --dropout 0.8 --layers 2 --epochs 2000 --conv lin

python main.py --data citeseer --lr 1e-3 --dropout 0.8 --layers 2 --epochs 2000 --conv lin

python main.py --data amac --lr 1e-2 --dropout 0.2 --layers 1 --epochs 400 --conv gcn

python main.py --data amap --lr 1e-2 --dropout 0.2 --layers 1 --epochs 400 --conv gcn

Note: It is undoubtedly extremely difficult to achieve the best results for all downstream tasks (feature reconstruction, node classification, node clustering) with unified hyperparameters ($\lambda_1$, $\lambda_2$, $l$, $\epsilon$ and $\alpha$). Therefore, different downstream tasks have different requirements for hyperparameters. If you want to reproduce the results in our paper, please refer to Section IV.G.

Contcat

This code has not been thoroughly verified and is only for learning and communication purposes. Please feel free to raise any questions or suggest better solutions by contacting limengran1998@163.com.