项目文件夹

文件

T

文件历史

Mufei Li 3f6f694159 [Bug fix] Various fix from bug bash (#3133 )

* Update

* Update

* Update dependencies

* Update

* Update

* Fix ogbn-products gat

* Update

* Update

* Reformat

* Fix typo in node2vec_random_walk

* Specify file encoding

* Working for 6.7

* Update

* Fix subgraph

* Fix doc for sample_neighbors_biased

* Fix hyperlink

* Add example for udf cross reducer

* Fix

* Add example for slice_batch

* Replace dgl.bipartite

* Fix GATConv

* Fix math rendering

* Fix doc

Co-authored-by: Ubuntu <ubuntu@ip-172-31-28-17.us-west-2.compute.internal>
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>
Co-authored-by: Ubuntu <ubuntu@ip-172-31-22-156.us-west-2.compute.internal>

2021-07-15 09:17:15 +08:00

main.py

[Refactor][Graph] Merge DGLGraph and DGLHeteroGraph (#1862 )

2020-07-28 14:30:41 +08:00

model_hetero.py

Update (#2057 )

2020-08-19 14:24:39 +08:00

model.py

[Example] Fix HAN (#1790 )

2020-07-11 17:29:19 +08:00

README.md

[Bug fix] Various fix from bug bash (#3133 )

2021-07-15 09:17:15 +08:00

train_sampling.py

add HAN sampling demo (#3005 )

2021-06-23 20:27:18 +08:00

utils.py

Update (#2057 )

2020-08-19 14:24:39 +08:00

README.md

Heterogeneous Graph Attention Network (HAN) with DGL

This is an attempt to implement HAN with DGL's latest APIs for heterogeneous graphs. The authors' implementation can be found here.

Usage

python main.py for reproducing HAN's work on their dataset.

python main.py --hetero for reproducing HAN's work on DGL's own dataset from here. The dataset is noisy because there are same author occurring multiple times as different nodes.

For sampling-based training, python train_sampling.py

Performance

Reference performance numbers for the ACM dataset:

	micro f1 score	macro f1 score
Paper	89.22	89.40
DGL	88.99	89.02
Softmax regression (own dataset)	89.66	89.62
DGL (own dataset)	91.51	91.66

We ran a softmax regression to check the easiness of our own dataset. HAN did show some improvements.