/graph

pytorch implementation of graph models, including gcn, gat, graphsage etc.

Primary LanguagePython

Introdcution

In this repo, we implement some common graph convolutional neural network layers (GCN, GAT, GraphSAGE...)

And in layers, we have two branchs: pyg and pytorch. Both ways implemented convolutional layers.

Usage

git clone https://github.com/downeykking/graph.git
cd directory gat or gcn
run main.py

Performances：

model GCN GAT
epoch 200 5000
learning rate 0.01 0.005
dropout 0.5 0.6
weight decay 5e-4 5e-4
hidden 16 8
seed 2022 2022
head / 8
alpha / 0.2
epoch time 0.0031s 0.0198s
total time 0.6171s 199.6522s
loss 0.4107 0.5054
test acc 82.60 84.40

Requirements：

torch=1.10.0+cu102
torch-geometric=2.0.2
pandas=1.3.4
numpy=1.21.4