dmlc--dgl
HQ
684a61ad09
* diffpool original file added * make diffpool fuse up and running * minor tweak on tu dataset statistics method * fix tu * break * delete break * pre_org * diffpool fuse reorg * fix random shuffling * fix bn * add dgl layers * early stopping * add readme * fix * add diffpool preprocess script * tweak tu dataset * tweak * tweak * tweak * tweak * tweak * preprocess dataset * fix early stopping * fix * fix * fix * tweak * readme * code review * code review * dataset code review * update README * code review * tu doc
30 行
795 B
Markdown
30 行
795 B
Markdown
Hierarchical Graph Representation Learning with Differentiable Pooling
|
|
============
|
|
|
|
|
|
Paper link: [https://arxiv.org/abs/1806.08804](https://arxiv.org/abs/1806.08804)
|
|
|
|
Author's code repo: [https://github.com/RexYing/diffpool](https://github.com/RexYing/diffpool)
|
|
|
|
This folder contains a DGL implementation of the DiffPool model. The first pooling layer is computed with DGL, and following pooling layers are computed with tensorized operation since the pooled graphs are dense.
|
|
|
|
Dependencies
|
|
------------
|
|
* PyTorch 1.0+
|
|
|
|
How to run
|
|
----------
|
|
|
|
```bash
|
|
python train.py --dataset ENZYMES --pool_ratio 0.10 --num_pool 1
|
|
python train.py --dataset DD --pool_ratio 0.15 --num_pool 1
|
|
```
|
|
Performance
|
|
-----------
|
|
ENZYMES 63.33% (with early stopping)
|
|
DD 79.31% (with early stopping)
|
|
|
|
|
|
## Dependencies
|
|
|