Distributed DeepGCNs

English | 简体中文

This is a distributed DeepGCNs model implemented and improved by High-Flyer AI. It can use multiple GPUs to achieve training acceleration.

DeepGCNs borrow concepts from CNNs, and mainly adapt residual/dense connections and dilated convolutions to GCN architectures.

Reference papers:

DeepGCNs (ICCV'2019, TPAMI'2021)
DeeperGCN (Arxiv'2020)
GNN'1000 (ICML'2021)

Requirements

Training

The raw data is from the public dataset, Open Graph Benchmark , which is integrated into the high-flyer dataset warehouse, hfai.datasets.

Node Property Prediction (ogbn-proteins)

submit the task to Yinghuo HPC:
```
 hfai python ogbn_proteins/main.py -- -n 1
```
run locally:
```
 python ogbn_proteins/main.py
```
Link Property Prediction (ogbl-collab)

submit the task to Yinghuo HPC:
```
 hfai python ogbl_collab/main.py -- -n 1
```
run locally:
```
 python ogbl_collab/main.py
```
Graph Property Prediction (ogbg-ppa)

submit the task to Yinghuo HPC:
```
 hfai python ogbg_ppa/main.py -- -n 1
```
run locally:
```
 python ogbg_ppa/main.py
```