【ICML-2019 SAGPooling】Self-Attention Graph Pooling
1.实验参数
Parameter | Value |
---|---|
Batch size | 128 |
Dataset | 可选: DD、MUTAG、NCI1、NCI109、PROTEINS, etc |
Dropout ratio | 0.5 |
Epochs | 10000 |
Exp name | 自命名: DD_Glo、MUTAG_Hie, etc |
Gpu index | 0 |
Hid | 128 |
Lr | 0.0005 |
Model | 可选: SAGPooling_Global、SAGPooling_Hierarchical |
Patience | 40 |
Pooling ratio | 0.5 |
Seed | 16 |
Test batch size | 1 |
Weight decay | 0.0001 |
2.运行程序
模型:SAGPooling_Global
数据集:DD
python main.py --exp_name=DD_Glo --dataset=DD --model=SAGPooling_Global
模型:SAGPooling_Hierarchical
数据集:PROTEINS
python main.py --exp_name=PROTEINS_Hie --dataset=PROTEINS --model=SAGPooling_Hierarchical
3.实验结果(8:1:1划分数据集,只做了一次实验的准确率,保留两位小数)
DD | MUTAG | NCI1 | NCI109 | PROTEINS | |
---|---|---|---|---|---|
SAGPooling_Global | 73.11 | 80.00 | 69.10 | 74.40 | 73.21 |
SAGPooling_Hierarchical | 67.23 | 70.00 | 66.18 | 70.77 | 69.64 |