SoGCN: Second-Order Graph Convolutional Networks

Last update: Aug 16, 2022

Overview

SoGCN: Second-Order Graph Convolutional Networks

This is the authors' implementation of paper "SoGCN: Second-Order Graph Convolutional Networks" in PyTorch. All the hyper-parameters and experiment settings have been included in this repo.

Requirements

For the GNN benchmarking part, our experiments are based on GNN Benchmark. Please follow the instructions in Benchmark Installation to install the running environment. Our code is tested with PyTorch 1.3.1 + Cuda Toolkit 10.0.

For the experiments on OGB Open Graph Benchmark, we built our models based on offical code. Please follow the instructions in Getting Started to configure the running environment. Our code is tested with PyTorch 1.4.0 + Cuda Toolkit 10.1.

Our experiments is conducted on a 4-core Nvidia Quadro P6000 GPU running on Ubuntu 18.04.2 LTS.

Reproduce Results

For SGS and GNN benchmark datasets, we provide a script named 'scripts/exp.py' to run a series of model training in screen sessions. You can type python scripts/exp.py -h to view its usage. To OGB benchmark dataset, we provide shell scripts 'scripts/run_ogbn_proteins.sh' and 'scripts/run_ogbg_molhiv.sh' to reproduce results with our hyper-parameter settings.

For convenience, below presents the commands to reproduce our results.

Synthetic Graph Spectrum Dataset

To train models on SGS datasets, run the following commands:

## High-Pass
python scripts/exp.py -a start -e highpass_sogcn -t SGS -g 1111 --dataset SGS_HIGH_PASS --config 'configs/SGS_node_regression_SoGCN.json'
python scripts/exp.py -a start -e highpass_sogcn -t SGS -g 1111 --dataset SGS_HIGH_PASS --config 'configs/SGS_node_regression_GCN.json'
python scripts/exp.py -a start -e highpass_sogcn -t SGS -g 1111 --dataset SGS_HIGH_PASS --config 'configs/SGS_node_regression_GIN.json'

## Low-Pass
python scripts/exp.py -a start -e lowpass_sogcn -t SGS -g 1111 --dataset SGS_LOW_PASS --config 'configs/SGS_node_regression_SoGCN.json'
python scripts/exp.py -a start -e lowpass_sogcn -t SGS -g 1111 --dataset SGS_LOW_PASS --config 'configs/SGS_node_regression_GCN.json'
python scripts/exp.py -a start -e lowpass_sogcn -t SGS -g 1111 --dataset SGS_LOW_PASS --config 'configs/SGS_node_regression_GIN.json'

## Band-Pass
python scripts/exp.py -a start -e bandpass_sogcn -t SGS -g 1111 --dataset SGS_BAND_PASS --config 'configs/SGS_node_regression_SoGCN.json'
python scripts/exp.py -a start -e bandpass_sogcn -t SGS -g 1111 --dataset SGS_BAND_PASS --config 'configs/SGS_node_regression_GCN.json'
python scripts/exp.py -a start -e bandpass_sogcn -t SGS -g 1111 --dataset SGS_BAND_PASS --config 'configs/SGS_node_regression_GIN.json'

Note the results will be saved to '_out/SGS_node_regression/'.

Open Graph Benchmarks

Running the following commands will reproduce our results on Open Graph Benchmark datasets:

scripts/run_ogbn_proteins.sh <log_dir> [<gpu_id>] [--test]
scripts/run_ogbg_molhiv.sh <log_dir> [<gpu_id>] [--test]

where log_dir specifies the folder to load or save output logs. The downloaded log files will be saved in _out/protein_nodeproppred and _out/molhiv_graphproppred for ogbn-protein and ogbn-molhiv datasets, respectively. gpu_id specifies the GPU device to run our models. Add --test if you only want to reload the log files and read out the testing results. The OGB dataset will be automatically downloaded into data/OGB directory at the first run.

To download the saved log files for ogb datasets, please run the following scripts:

bash scripts/download_logfiles_ogb.sh

GNN Benchmarks

To test on our pretrained models on GNN benchmarks, please follow the steps as below:

Download our pretrained models.

# make sure the commands below are executed in the root directory of this project
bash scripts/download_pretrained_molecules.sh
bash scripts/download_pretrained_superpixels.sh
bash scripts/download_pretrained_SBMs.sh

Pretrained models will be downloaded to '_out/molecules_graph_regression', '_out/superpixels_graph_classification', '_out/SBMs_node_classification', respectively.

Type the commands for different tasks

Molecules Graph Regression

## ZINC
python main_molecules_graph_regression.py --model SoGCN --dataset ZINC --gpu_id 0 --test True --out_dir _out/molecules_graph_regression/zinc_sogcn
python main_molecules_graph_regression.py --model SoGCN --dataset ZINC --gpu_id 0 --test True --out_dir _out/molecules_graph_regression/zinc_sogcn_gru

Superpixels Graph Classification

## MNIST
python main_superpixels_graph_classification.py --model SoGCN --dataset MNIST --gpu_id 0 --test True --out_dir _out/superpixels_graph_classification/mnist_sogcn
python main_superpixels_graph_classification.py --model SoGCN --dataset MNIST --gpu_id 0 --test True --out_dir _out/superpixels_graph_classification/mnist_sogcn_gru


## CIFAR10
python main_superpixels_graph_classification.py --model SoGCN --dataset CIFAR10 --gpu_id 0 --test True --out_dir _out/superpixels_graph_classification/cifar10_sogcn
python main_superpixels_graph_classification.py --model SoGCN --dataset CIFAR10 --gpu_id 0 --test True --out_dir _out/superpixels_graph_classification/cifar10_sogcn_gru

SBMs Node Classification

## CLUSTER
python main_SBMs_node_classification.py --model SoGCN --dataset SBM_CLUSTER  --verbose True --gpu_id 0 --test True --out_dir _out/SBMs_node_classification/cluster_sogcn
python main_SBMs_node_classification.py --model SoGCN --dataset SBM_CLUSTER  --verbose True --gpu_id 0 --test True --out_dir _out/SBMs_node_classification/cluster_sogcn_gru

## PATTERN
python main_SBMs_node_classification.py --model SoGCN --dataset SBM_PATTERN  --verbose True --gpu_id 0 --test True --out_dir _out/SBMs_node_classification/pattern_sogcn
python main_SBMs_node_classification.py --model SoGCN --dataset SBM_PATTERN  --verbose True --gpu_id 0 --test True --out_dir _out/SBMs_node_classification/pattern_sogcn_gru

SoGCN: Second-Order Graph Convolutional Networks

Related tags

Overview

SoGCN: Second-Order Graph Convolutional Networks

Requirements

Reproduce Results

Synthetic Graph Spectrum Dataset

Open Graph Benchmarks

GNN Benchmarks

Owner

Yuehao

Official implementation for "Style Transformer for Image Inversion and Editing" (CVPR 2022)

A PyTorch implementation: "LASAFT-Net-v2: Listen, Attend and Separate by Attentively aggregating Frequency Transformation"

This initial strategy was developed specifically for larger pools and is based on taking a moving average and deriving Bollinger Bands to create a projected active liquidity range.

Code of Adverse Weather Image Translation with Asymmetric and Uncertainty aware GAN

In-place Parallel Super Scalar Samplesort (IPS⁴o)

Codes for our IJCAI21 paper: Dialogue Discourse-Aware Graph Model and Data Augmentation for Meeting Summarization

Sudoku solver - A sudoku solver with python

This repository contains the code for the paper in EMNLP 2021: "HRKD: Hierarchical Relational Knowledge Distillation for Cross-domain Language Model Compression".

Implementation for the "Surface Reconstruction from 3D Line Segments" paper.

Code release for the ICML 2021 paper "PixelTransformer: Sample Conditioned Signal Generation".

Ray tracing of a Schwarzschild black hole written entirely in TensorFlow.

SpinalNet: Deep Neural Network with Gradual Input

The official implementation of the research paper "DAG Amendment for Inverse Control of Parametric Shapes"

Source code for "OmniPhotos: Casual 360° VR Photography"

MLOps will help you to understand how to build a Continuous Integration and Continuous Delivery pipeline for an ML/AI project.

BBScan py3 - BBScan py3 With Python

PixelPyramids: Exact Inference Models from Lossless Image Pyramids (ICCV 2021)

[ECCV'20] Convolutional Occupancy Networks

PromptDet: Expand Your Detector Vocabulary with Uncurated Images

Official implementation for "QS-Attn: Query-Selected Attention for Contrastive Learning in I2I Translation" (CVPR 2022)