PAS-OGB

We apply our novel framework PAS to automatically learn data-specific pooling architectures for graph classification task.

Requirements

Python>=3.7 Pytorch==1.8.0 pytorch_geometric==2.0.1 numpy==1.21.2 pandas==1.3.4 
scikit-learn==0.24.2 ogb>=1.3.2 deep_gcns_torch LibAUC 

Results on OGB

ogbg-molhiv dataset

Dataset	Method	Test AUC	Validation AUC	#Parameters	Hardware
ogbg-molhiv	PAS	0.8221 ± 0.0017	0.8178 ± 0.0031	26,706,952	RTX3090
ogbg-molhiv	PAS+FingerPrint	0.8420 ± 0.0015	0.8238 ± 0.0028	26,706,953	RTX3090

ogbg-molpcba dataset

Dataset	Method	Test AP	Validation AP	Hardware
ogbg-molpcba	PAS	0.3012 ± 0.0039	0.3151 ± 0.0047	RTX3090

ogbg-molppa dataset

Dataset	Method	Test ACC	Validation ACC	Hardware
ogbg-ppa	PAS	0.7828 ± 0.0024	0.7523 ± 0.0028	RTX3090
ogbg-ppa	PAS+F2GNN(hidden_size 128)	0.7842 ± 0.0023	0.7373 ± 0.0021	RTX3090
ogbg-ppa	PAS+F2GNN(hidden_size 512)	0.8201 ± 0.0019	0.7720 ± 0.0023	RTX3090

Training Process for ogbg-molhiv

1. Search Architecture.

python hiv_train_search.py --gpu 0 --num_layers 14 --epochs 50 --data ogbg-molhiv
--remove_pooling True

2. Extract finerprints and train Random Forest.

python extract_fingerprint.py
python random_forest.py

3. Finetune the model.

python -u finetune.py --data ogbg-molhiv --gpu 0 --dropout 0.2 --lr 0.1 
--batch_size 256 --gamma 700 --epochs 400 --hidden_size 512 
--arch_filename ./exp_res/ogbg-molhiv-searched_res-20220120-220405-eps0.0-reg1e-05.txt

If you want to use the model framework you searched for, please enter your model address after --arch_filename

4. Finetune the model with FingerPrints, the FT model can be found in the release, its name is BS_256-NF_full_valid_best_AUC-FP_E_341_R0.pth. Create the folder model_0206_gamma_500 and drage the model file into it.

python -u finetune_Drop.py --data ogbg-molhiv --gpu 3 --dropout 0.1 --lr 0.005 --batch_size 256 --gamma 100 --epochs 40 --hidden_size 512 --arch_filename ./exp_res/ogbg-molhiv-searched_res-20220120-220405-eps0.0-reg1e-05.txt

Training Process for ogbg-molpcba

1. Search Architecture.

python model_search.py --gpu 0 --num_layers 5 --epochs 5 --data ogbg-molpcba
--remove_pooling True

2. Finetune the model.

python finetune.py --gpu 0 --dropout 0.5 --lr 0.001 --batch_size 100 --num_layers 5 --epochs 100 --hidden_size 384  --arch_filename ./exp_res/ogbg-molpcba-searched_res-20220316-235183-eps0.0-reg1e-05.txt

Training Process for ogbg-ppa(PAS)

1. Search Architecture

python model_search.py --gpu 0 --num_layers 3 --epochs 5 --data ogbg-ppa
--remove_pooling True

2. Finetune the model.

python finetune.py --gpu 1 --dropout 0.5 --lr 0.0005 --batch_size 24 --num_layers 3 --epochs 200 --hidden_size 512 --arch_filename ./exp_res/ogbg-ppa-searched_res-20220217-214538-eps0.0-reg1e-05.txt

Training Process for ogbg-ppa(PAS+F2GNN)

1. Search Architecture

python model_search.py --gpu 0 --num_layers 5 --epochs 10 --batch_size 24 --hidden_size 64 --data ogbg-ppa
--remove_pooling True

2. Finetune the model.

python finetune.py --gpu 1 --dropout 0.5 --lr 0.0005 --batch_size 24 --num_layers 5 --epochs 200 --warmup_epochs 20 --hidden_size 512 --arch_filename ./exp_res/ogbg-ppa-searched_res-20220415-112841-eps0.0-reg1e-05.tx