GraphLand Baselines

This repository provides the code for reproducing the results of graph foundation models and tabular baselines in the paper GraphLand: Evaluating Graph Machine Learning Models on Diverse Industrial Data.

Code

In source directory, one can find the source code for reproducing experiments in our paper. Each subdirectory is adapted from the corresponding open source repository:

• gbdt — github.com/yandex-research/tabular-dl-tabr
• anygraph — github.com/HKUDS/AnyGraph
• opengraph — github.com/HKUDS/OpenGraph
• tsgnn — github.com/benfinkelshtein/EquivarianceEverywhere
• gcope — github.com/cshhzhao/GCOPE

Environment

To install the enviroment suitable for all the considered baselines, do the following:

1. Install pixi according to https://pixi.sh/dev/installation/

2. Run pixi install to install the dependencies listed in pixi.toml

Datasets

1. Download GraphLand datasets and unzip them into datasets directory

2. Run pixi run python scripts/nfa.py to prepare NFA features

3. Run pixi run python scripts/convert.py to convert datasets into the format required by each baseline

Experiments

1. Change working directory to the specific baseline repository:

# for GBDT
cd source/gbdt

# for AnyGraph
cd source/anygraph/node_classification

# for OpenGraph
cd source/opengraph/node_classification

# for TS-GNN
cd source/tsgnn

# for GCOPE
cd source/gcope

2. Run the script:

pixi run python __run.py