AutoGraph

Introduction

This is the official repo for hosting AutoGraph challenge datasets/code from KDD Cup 2020, targeting automated node classification task. To that end, we provide in this repo 14 datasets from public and private sources to fully evaluate the generalization ability of AutoML algorithms.

Running locally

We provide full details for evaluating locally. Please download the datasets and put under the folder src/data. Then pull the prepared docker nehzux/kddcup2020:v2 and run command docker run --gpus=0 -it --rm -v "$(pwd):/app/autograph" -w /app/autograph nehzux/kddcup2020:v2

Go to the directory of run_local_test.py and run python run_local_test.py --dataset_dir=data/a --code_dir=code_submission/0_baseline for testing baseline on dataset a.

Dataset license:

We include 14 datasets in total here in this repository. 10 datasets come from public datasets and 4 come from 4paradigm. We conclude the licenses for each dataset below. If you find anything inappropriate and have any question regrading the license, please contact us by: xuzhen[AT]4paradigm[DOT]com

Dataset ID	Origin Source	License
a	Cora	MIT
b	CiteSeer	CC BY-NC-SA 3.0
c	Reddit	MIT
d	20 Newsgroups	Credit to Ken Lang and Tom Mitchell
e	4paradigm	CC BY-NC 4.0
f	Amazon Computer	MIT
g	Coauthor Physics	MIT
h	ohsumed	CC BY-NC 4.0
i	4paradigm	CC BY-NC 4.0
k	Amazon Photo	MIT
l	Coauthor CS	MIT
m	R8	Copyright to Reuters Ltd
n	4paradigm	CC BY-NC 4.0
o	4paradigm	CC BY-NC 4.0

References:

• Bridging the Gap of AutoGraph between Academia and Industry: Analysing AutoGraph Challenge at KDD Cup 2020
• https://github.com/aister2020/KDDCUP_2020_AutoGraph_1st_Place
• https://github.com/Unkrible/AutoGraph2020
• https://github.com/white-bird/kdd2020_GCN