Source code for From Stars to Subgraphs

Related tags

Deep LearningGNNAsKernel
Overview

GNNAsKernel

Official code for From Stars to Subgraphs: Uplifting Any GNN with Local Structure Awareness

Visualizations

GNN-AK(+)

GNN-AK

GNN-AK(+) with SubgraphDrop

GNN-AK-S: GNN-AK with SubgraphDrop

Setup

# params
# 10/6/2021, newest packages. 
ENV=gnn_ak
CUDA=11.1
TORCH=1.9.1
PYG=2.0.1

# create env 
conda create --name $ENV python=3.9 -y
conda activate $ENV

# install pytorch 
conda install pytorch=$TORCH torchvision torchaudio cudatoolkit=$cuda -c pytorch -c nvidia -y

# install pyg2.0
conda install pyg=$PYG -c pyg -c conda-forge -y

# install ogb 
pip install ogb

# install rdkit
conda install -c conda-forge rdkit -y

# update yacs and tensorboard
pip install yacs==0.1.8 --force  # PyG currently use 0.1.6 which doesn't support None argument. 
pip install tensorboard
pip install matplotlib

Code structure

core/ contains all source code.
train/ contains all scripts for available datasets.

  • Subgraph extraction is implemented as data transform operator in PyG. See core/transform.py. The transform layer will built the mapping from original nodes and edges to all subgraphs.
  • The mappings are used directly in GNN-AK(+) to online build the combined subgraphs for each graph, see core/model.py. (For each graph with N node, N subgraphs are combined to a gaint subgraph first. Then for batch, all combined gaint subgraphs are combined again.)
  • SubgraphDrop is implemented inside core/transform.py, see here. And the usage in core/model.py.
  • core/model_utils/pyg_gnn_wrapper.py is the place to add your self-designed GNN layer X and then use X-AK(+) on fly~

Hyperparameters

See core/config.py for all options.

Run normal GNNs

See core/model_utls/pyg_gnn_wrapper.py for more options.

Custom new GNN convolutional layer 'X' can be plugged in core/model_utls/pyg_gnn_wrapper.py, and use 'X' as model.gnn_type option.

# Run different normal GNNs 
python -m train.zinc model.mini_layers 0 model.gnn_type GINEConv
python -m train.zinc model.mini_layers 0 model.gnn_type SimplifiedPNAConv
python -m train.zinc model.mini_layers 0 model.gnn_type GCNConv
python -m train.zinc model.mini_layers 0 model.gnn_type GATConv
python -m train.zinc model.mini_layers 0 model.gnn_type ...

python -m train.zinc model.num_layers 6 model.mini_layers 0 model.gnn_type GCNConv # 6-layer GCN

Run different datasets

See all available datasets under train folder.

# Run different datasets
python -m train.zinc 
python -m train.cifar10 
python -m train.counting 
python -m train.graph_property 
python -m ...

Run GNN-AK

Fully theoretically explained by Subgraph-1-WL*.

Use: model.mini_layers 1 (or >1) model.embs "(0,1)" model.hops_dim 0 

python -m train.zinc model.mini_layers 1 model.gnn_type GINEConv model.embs "(0,1)" model.hops_dim 0

Run GNN-AK+

At least as powerful as GNN-AK (or more powerful).

Use: model.mini_layers 1 (or >1) model.embs "(0,1,2)" model.hops_dim 16
These are set as default. See core/config.py.

# Run GNN-AK+ with different normal GNNs
python -m train.zinc model.mini_layers 1 model.gnn_type GINEConv            # 1-layer base model
python -m train.zinc model.mini_layers 1 model.gnn_type SimplifiedPNAConv   # 1-layer base model
python -m train.zinc model.mini_layers 2 model.gnn_type GINEConv            # 2-layer base model
python -m train.zinc model.mini_layers 2 model.gnn_type SimplifiedPNAConv   # 2-layer base model

Run with different number of GNN-AK(+) iterations

Changing number of outer layers.

python -m train.zinc model.num_layers 4 
python -m train.zinc model.num_layers 6 
python -m train.zinc model.num_layers 8

Run with different subgraph patterns

See core/transform.py for detailed implementation.

python -m train.zinc subgraph.hops 2      # 2-hop egonet
python -m train.zinc subgraph.hops 3      # 3-hop egonet

# Run with random-walk subgraphs based on node2vec 
python -m train.zinc subgraph.hops 0 subgraph.walk_length 10 subgraph.walk_p 1.0 subgraph.walk_q 1.0

Run GNN-AK(+) with SubgraphDrop

See option sampling section under core/config.py.

Change sampling.redundancy(R in the paper) to change the resource usage.

python -m train.zinc sampling.mode shortest_path sampling.redundancy 1 sampling.stride 5 sampling.batch_factor 4
python -m train.zinc sampling.mode shortest_path sampling.redundancy 3 sampling.stride 5 sampling.batch_factor 4
python -m train.zinc sampling.mode shortest_path sampling.redundancy 5 sampling.stride 5 sampling.batch_factor 4


python -m train.cifar10 sampling.mode random sampling.redundancy 1 sampling.random_rate 0.07 sampling.batch_factor 8 
python -m train.cifar10 sampling.mode random sampling.redundancy 3 sampling.random_rate 0.21 sampling.batch_factor 8 
python -m train.cifar10 sampling.mode random sampling.redundancy 5 sampling.random_rate 0.35 sampling.batch_factor 8 
## Note: sampling.random_rate = 0.07*sampling.redundancy. 0.07 is set based on dataset.

Results

GNN-AK boosts expressiveness

GNN-AK boosts expressiveness

GNN-AK boosts practical performance

GNN-AK boosts practical performance

Cite

Please cite our work if you use our code!

@inproceedings{
anonymous2022from,
title={From Stars to Subgraphs: Uplifting Any {GNN} with Local Structure Awareness},
author={Anonymous},
booktitle={Submitted to The Tenth International Conference on Learning Representations },
year={2022},
url={https://openreview.net/forum?id=Mspk_WYKoEH},
note={under review}
}
Owner
GitHub Repository
A PyTorch implementation of the paper "Semantic Image Synthesis via Adversarial Learning" in ICCV 2017

Semantic Image Synthesis via Adversarial Learning This is a PyTorch implementation of the paper Semantic Image Synthesis via Adversarial Learning. Req

Seonghyeon Nam 146 Nov 25, 2022
Repository for self-supervised landmark discovery

self-supervised-landmarks Repository for self-supervised landmark discovery Requirements pytorch pynrrd (for 3d images) Usage The use of this models i

Riddhish Bhalodia 2 Apr 18, 2022
A simple code to perform canny edge contrast detection on images.

CECED-Canny-Edge-Contrast-Enhanced-Detection A simple code to perform canny edge contrast detection on images. A simple code to process images using c

Happy N. Monday 3 Feb 15, 2022
Pytorch Implementation of rpautrat/SuperPoint

SuperPoint-Pytorch (A Pure Pytorch Implementation) SuperPoint: Self-Supervised Interest Point Detection and Description Thanks This work is based on:

76 Dec 27, 2022
Meshed-Memory Transformer for Image Captioning. CVPR 2020

M²: Meshed-Memory Transformer This repository contains the reference code for the paper Meshed-Memory Transformer for Image Captioning (CVPR 2020). Pl

AImageLab 422 Dec 28, 2022
A texturizer that I just made. Nothing special here.

texturizer This is a little project that I did with an hour's time. It texturizes an image given a image and a texture to texturize it with. There is

1 Nov 11, 2021
Code for SyncTwin: Treatment Effect Estimation with Longitudinal Outcomes (NeurIPS 2021)

SyncTwin: Treatment Effect Estimation with Longitudinal Outcomes (NeurIPS 2021) SyncTwin is a treatment effect estimation method tailored for observat

Zhaozhi Qian 3 Nov 03, 2022
PN-Net a neural field-based framework for depth estimation from single-view RGB images.

PN-Net We present a neural field-based framework for depth estimation from single-view RGB images. Rather than representing a 2D depth map as a single

1 Oct 02, 2021
Deep deconfounded recommender (Deep-Deconf) for paper "Deep causal reasoning for recommendations"

Deep Causal Reasoning for Recommender Systems The codes are associated with the following paper: Deep Causal Reasoning for Recommendations, Yaochen Zh

Yaochen Zhu 22 Oct 15, 2022
All materials of Cassandra Event, Udyam'22

Cassandra 2022 Workspace Workshop Materials Workshop-1 Workshop-2 Workshop-3 Workshop-4 Assignments Assignment-1 Assignment-2 Assignment-3 Resources P

36 Dec 31, 2022
[TIP 2021] SADRNet: Self-Aligned Dual Face Regression Networks for Robust 3D Dense Face Alignment and Reconstruction

SADRNet Paper link: SADRNet: Self-Aligned Dual Face Regression Networks for Robust 3D Dense Face Alignment and Reconstruction Requirements python

Multimedia Computing Group, Nanjing University 99 Dec 30, 2022
Proof of concept GnuCash Webinterface

Proof of Concept GnuCash Webinterface This may one day be a something truly great. Milestones [ ] Browse accounts and view transactions [ ] Record sim

Josh 14 Dec 28, 2022
PyTorch implementation of Higher Order Recurrent Space-Time Transformer

Higher Order Recurrent Space-Time Transformer (HORST) This is the official PyTorch implementation of Higher Order Recurrent Space-Time Transformer. Th

13 Oct 18, 2022
STBP is a way to train SNN with datasets by Backward propagation.

Spiking neural network (SNN), compared with depth neural network (DNN), has faster processing speed, lower energy consumption and more biological interpretability, which is expected to approach Stron

Ling Zhang 18 Dec 09, 2022
Monitora la qualità della ricezione dei segnali radio nelle province siciliane.

FMap-server Monitora la qualità della ricezione dei segnali radio nelle province siciliane. Conversion data Frequency - StationName maps are stored in

Triglie 5 May 24, 2021
Deep Illuminator is a data augmentation tool designed for image relighting. It can be used to easily and efficiently generate a wide range of illumination variants of a single image.

Deep Illuminator Deep Illuminator is a data augmentation tool designed for image relighting. It can be used to easily and efficiently generate a wide

George Chogovadze 52 Nov 29, 2022
TorchGeo is a PyTorch domain library, similar to torchvision, that provides datasets, transforms, samplers, and pre-trained models specific to geospatial data.

TorchGeo is a PyTorch domain library, similar to torchvision, that provides datasets, transforms, samplers, and pre-trained models specific to geospatial data.

Microsoft 1.3k Dec 30, 2022
Multi-Horizon-Forecasting-for-Limit-Order-Books

Multi-Horizon-Forecasting-for-Limit-Order-Books This jupyter notebook is used to demonstrate our work, Multi-Horizon Forecasting for Limit Order Books

Zihao Zhang 116 Dec 23, 2022
Weakly-supervised semantic image segmentation with CNNs using point supervision

Code for our ECCV paper What's the Point: Semantic Segmentation with Point Supervision. Summary This library is a custom build of Caffe for semantic i

27 Sep 14, 2022
Implementation of "Bidirectional Projection Network for Cross Dimension Scene Understanding" CVPR 2021 (Oral)

Bidirectional Projection Network for Cross Dimension Scene Understanding CVPR 2021 (Oral) [ Project Webpage ] [ arXiv ] [ Video ] Existing segmentatio

Hu Wenbo 135 Dec 26, 2022