Giter Site home page Giter Site logo

signnap-master-online's Introduction

SIGNNAP-master-online

This is a Pytorch implementation of our paper "Learning Node Representations against Perturbations"

1. Datasets:
We use six benchmarks: Pubmed, Facbook, Coauthor-CS, Coauthor-Phy. Amazon-Computer and Amazon-Photo The transductive learning setting is conducted on Pubmed, Facbook, Coauthor-CS, Amazon-Computer, Amazon-Photo, and the inductive learning setting is conducted on Coauthor-Phy.

2. Requirements:
(1) GPU machine
(2) Necessary packages:
python3.5; pytorch=1.4.0; tqdm=4.27.0; tensorboardX=1.8; pandas=0.25; numpy=1.15; networkx=2.2; logger=1.4; scipy=1.1; scikit-learn=0.20

3. Notes:
(1) Note that some of the codes may be redundant and are not useful, we will polish them later.
(2) Note that some of the codes are based on different released codes, so the same arguement in different .py files may have different meanings.
(3) Note that the following comands consist of both training and evaluation, if you do not want to train Ours, you can directly use the trained models we provided in xxxx/src/saved_models/.
Then, you can directly check the ############################### Evaluation for Ours ######################################## part and run the evaluation results.
For example:
--> CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=pubmed --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1
--> CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=facebook_page --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1
--> CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_cs --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1
--> CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_phy --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1
--> CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=amazon_computer --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0
--> CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_phy --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

4. We will illustrate how to run the codes as following.

In order to run our codes, you need to get in the: xxxx/src/ file first.
(1) On Pubmed dataset
############################### Baselines ########################################
GCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=pubmed --sampling_percent=1.0 --type=multigcn --nbaseblocklayer=0

ResGCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=pubmed --sampling_percent=1.0 --type=resgcn --nbaseblocklayer=1

JKNet: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=pubmed --sampling_percent=1.0 --type=densegcn --nbaseblocklayer=1

GraphSage: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=pubmed --sampling_percent=1.0 --nhiddenlayer=2

GAT: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=pubmed --sampling_percent=1.0 --nhiddenlayer=1

DropEdge(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=pubmed --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0

DropEdge(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=pubmed --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1

DropEdge(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=pubmed --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1

DropEdge(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=pubmed --sampling_percent=0.7 --nhiddenlayer=2

DropEdge(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=pubmed --sampling_percent=0.7 --nhiddenlayer=1

DGI: CUDA_VISIBLE_DEVICES=your_gpu_num python train_DGI.py --dataset=pubmed

############################### Ours ########################################
Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=pubmed --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=pubmed --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=pubmed --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_graphsage.py --dataset=pubmed --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_gat.py --dataset=pubmed --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1

############################### Evaluation for Ours ########################################
Since our model is an unsupervised learning one, we provide the following code for evaluation after the model training.

Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=pubmed --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=pubmed --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=pubmed --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_graphsage.py --dataset=pubmed --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_gat.py --dataset=pubmed --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1

(2) On Facebook dataset
############################### Baselines ########################################
GCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=facebook_page --sampling_percent=1.0 --type=multigcn --nbaseblocklayer=0

ResGCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=facebook_page --sampling_percent=1.0 --type=resgcn --nbaseblocklayer=1

JKNet: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=facebook_page --sampling_percent=1.0 --type=densegcn --nbaseblocklayer=1

GraphSage: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=facebook_page --sampling_percent=1.0 --nhiddenlayer=2

GAT: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=facebook_page --sampling_percent=1.0 --nhiddenlayer=1

DropEdge(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=facebook_page --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0

DropEdge(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=facebook_page --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1

DropEdge(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=facebook_page --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1

DropEdge(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=facebook_page --sampling_percent=0.7 --nhiddenlayer=2

DropEdge(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=facebook_page --sampling_percent=0.7 --nhiddenlayer=1

DGI: CUDA_VISIBLE_DEVICES=your_gpu_num python train_DGI.py --dataset=facebook_page

############################### Ours ########################################
Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=facebook_page --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=facebook_page --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=facebook_page --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_graphsage.py --dataset=facebook_page --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_gat.py --dataset=facebook_page --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1

############################### Evaluation for Ours ########################################
Since our model is an unsupervised learning one, we provide the following code for evaluation after the model training.

Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=facebook_page --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=facebook_page --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=facebook_page --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_graphsage.py --dataset=facebook_page --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_gat.py --dataset=facebook_page --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1

(3) On Coauthor-CS dataset
############################### Baselines ########################################
GCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_cs --sampling_percent=1.0 --type=multigcn --nbaseblocklayer=0

ResGCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_cs --sampling_percent=1.0 --type=resgcn --nbaseblocklayer=1

JKNet: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_cs --sampling_percent=1.0 --type=densegcn --nbaseblocklayer=1

GraphSage: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=coauthor_cs --sampling_percent=1.0 --nhiddenlayer=2

GAT: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=coauthor_cs --sampling_percent=1.0 --nhiddenlayer=1

DropEdge(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_cs --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0

DropEdge(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_cs --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1

DropEdge(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_cs --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1

DropEdge(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=coauthor_cs --sampling_percent=0.7 --nhiddenlayer=2

DropEdge(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=coauthor_cs --sampling_percent=0.7 --nhiddenlayer=1

DGI: CUDA_VISIBLE_DEVICES=your_gpu_num python train_DGI.py --dataset=coauthor_cs

############################### Ours ########################################
Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=coauthor_cs --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=coauthor_cs --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=coauthor_cs --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_graphsage.py --dataset=coauthor_cs --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_gat.py --dataset=coauthor_cs --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1

############################### Evaluation for Ours ########################################
Since our model is an unsupervised learning one, we provide the following code for evaluation after the model training.

Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_cs --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_cs --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_cs --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_graphsage.py --dataset=coauthor_cs --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_gat.py --dataset=coauthor_cs --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1

(4) On Coauthor-Phy dataset
############################### Baselines ########################################
GCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_phy --sampling_percent=1.0 --type=multigcn --nbaseblocklayer=0

ResGCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_phy --sampling_percent=1.0 --type=resgcn --nbaseblocklayer=1

JKNet: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_phy --sampling_percent=1.0 --type=densegcn --nbaseblocklayer=1

GraphSage: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=coauthor_phy --sampling_percent=1.0 --nhiddenlayer=2

GAT: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=coauthor_phy --sampling_percent=1.0 --nhiddenlayer=1

DropEdge(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_phy --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0

DropEdge(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_phy --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1

DropEdge(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=coauthor_phy --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1

DropEdge(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=coauthor_phy --sampling_percent=0.7 --nhiddenlayer=2

DropEdge(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=coauthor_phy --sampling_percent=0.7 --nhiddenlayer=1

DGI: CUDA_VISIBLE_DEVICES=your_gpu_num python train_DGI.py --dataset=coauthor_phy

############################### Ours ########################################
Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=coauthor_phy --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=coauthor_phy --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=coauthor_phy --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_graphsage.py --dataset=coauthor_phy --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_gat.py --dataset=coauthor_phy --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1

############################### Evaluation for Ours ########################################
Since our model is an unsupervised learning one, we provide the following code for evaluation after the model training.

Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_phy --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_phy --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=coauthor_phy --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_graphsage.py --dataset=coauthor_phy --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_gat.py --dataset=coauthor_phy --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1

(5) On Amazon-Com dataset
############################### Baselines ########################################
GCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_computer --sampling_percent=1.0 --type=multigcn --nbaseblocklayer=0 --weight_decay=0.0

ResGCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_computer --sampling_percent=1.0 --type=resgcn --nbaseblocklayer=1 --weight_decay=0.0

JKNet: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_computer --sampling_percent=1.0 --type=densegcn --nbaseblocklayer=1 --weight_decay=0.0

GraphSage: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=amazon_computer --sampling_percent=1.0 --nhiddenlayer=2 --weight_decay=0.0

GAT: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=amazon_computer --sampling_percent=1.0 --nhiddenlayer=1 --weight_decay=0.0

DropEdge(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_computer --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --weight_decay=0.0

DropEdge(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_computer --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --weight_decay=0.0

DropEdge(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_computer --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --weight_decay=0.0

DropEdge(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=amazon_computer --sampling_percent=0.7 --nhiddenlayer=2 --weight_decay=0.0

DropEdge(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=amazon_computer --sampling_percent=0.7 --nhiddenlayer=1 --weight_decay=0.0

DGI: CUDA_VISIBLE_DEVICES=your_gpu_num python train_DGI.py --dataset=amazon_computer --weight_decay=0.0

############################### Ours ########################################
Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=amazon_computer --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=amazon_computer --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=amazon_computer --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_graphsage.py --dataset=amazon_computer --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_gat.py --dataset=amazon_computer --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

############################### Evaluation for Ours ########################################
Since our model is an unsupervised learning one, we provide the following code for evaluation after the model training.

Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=amazon_computer --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=amazon_computer --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=amazon_computer --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_graphsage.py --dataset=amazon_computer --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_gat.py --dataset=amazon_computer --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

(6) On Amazon-Pho dataset
############################### Baselines ########################################
GCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_photo --sampling_percent=1.0 --type=multigcn --nbaseblocklayer=0 --weight_decay=0.0

ResGCN: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_photo --sampling_percent=1.0 --type=resgcn --nbaseblocklayer=1 --weight_decay=0.0

JKNet: CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_photo --sampling_percent=1.0 --type=densegcn --nbaseblocklayer=1 --weight_decay=0.0

GraphSage: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=amazon_photo --sampling_percent=1.0 --nhiddenlayer=2 --weight_decay=0.0

GAT: CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=amazon_photo --sampling_percent=1.0 --nhiddenlayer=1 --weight_decay=0.0

DropEdge(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_photo --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --weight_decay=0.0

DropEdge(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_photo --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --weight_decay=0.0

DropEdge(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_new.py --dataset=amazon_photo --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --weight_decay=0.0

DropEdge(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_graphsage.py --dataset=amazon_photo --sampling_percent=0.7 --nhiddenlayer=2 --weight_decay=0.0

DropEdge(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_news_gat.py --dataset=amazon_photo --sampling_percent=0.7 --nhiddenlayer=1 --weight_decay=0.0

DGI: CUDA_VISIBLE_DEVICES=your_gpu_num python train_DGI.py --dataset=amazon_photo --weight_decay=0.0

############################### Ours ########################################
Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=amazon_photo --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=amazon_photo --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours.py --dataset=amazon_photo --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_graphsage.py --dataset=amazon_photo --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python train_Ours_gat.py --dataset=amazon_photo --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

############################### Evaluation for Ours ########################################
Since our model is an unsupervised learning one, we provide the following code for evaluation after the model training.

Ours(GCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=amazon_photo --sampling_percent=0.7 --type=multigcn --nbaseblocklayer=0 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(ResGCN): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=amazon_photo --sampling_percent=0.7 --type=resgcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(JKNet): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervisd.py --dataset=amazon_photo --sampling_percent=0.7 --type=densegcn --nbaseblocklayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(GraphSage): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_graphsage.py --dataset=amazon_photo --sampling_percent=0.7 --nhiddenlayer=2 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

Ours(GAT): CUDA_VISIBLE_DEVICES=your_gpu_num python eval_unsupervised_gat.py --dataset=amazon_photo --sampling_percent=0.7 --nhiddenlayer=1 --nce_k=1024 --nce_t=0.1 --weight_decay=0.0

For codes of other methods such as DeepWalk, they have been provided by the authors and can be widely reached online, we do not repeat them here

Thanks for your interest in our paper. We benefit a lot from codes provided by other researchers, we would like to thank them here.

Also, this paper is under review now, we will provide more information later.

signnap-master-online's People

Contributors

xuchensjtu avatar

Stargazers

avatar Huangjie Zheng avatar avatar

Watchers

avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.