Implementation for CCA-AGC model (A Contrastive Learning Method with Cluster-preserving Augmentation for Attributed Graph Clustering).

Implementation

pretrain.py: pretrain multilevel contrast to get initial parameters and node representations.

train_conclu.py: jointly train the whole model.

Parameters Setting

Dataset	Encoding dimension	Projecting dimension	Activation Function	Learning rate	kNN	p_e	p_m	Epoch	T
Cora	512-256	1024	ReLu	0.0001	0	0.85	0.1	200	1
CiteSeer	1024-512	1024	PReLu	0.0005	1	0.65	0.4	300	1
PubMed	1024-512	512	ReLu	0.001	5	0.9	0.2	200	1
WikiCS	1024-1024	128	PReLu	0.01/0.005	0	0.01	0.2	200	20
AmazonCom	128-128	1024	PReLu	0.0005	10	0.65	0	200	200
Amazon-Photo	512-128	1024	ReLu	0.00003	6	0.85	0	200	20
Coauthor-CS	256-256	1024	PReLu	0.001	0	0.5	0	200	200

Example:

python train_conclu.py --dataset Cora --hidden 512 --out_dim 256 --pro_hid 1024 --activation relu --k 0 --rm 85 --mask 0.1 --lr 0.0001