---

• Pytorch Geometric Loader
• Baselines
  • Environments
  • Training
  • Performances:
• Datasets Overview
  • Summary

Pytorch Geometric Loader

Pytorch Geometric (>=2.2.0) provides an easy-to-use dataset loader for DGraphFin. Below is an example in Pytorch Geometric with only a few lines of code to load DGraphFin and get the train/valid/test mask.

import torch_geometric

# check your torch_geometric version and make sure it is not lower than 2.2.0
print(torch_geometric.__version__)
>>> '2.2.0'

# Please download DGraphFin dataset file 'DGraphFin.zip' on our website 'https://dgraph.xinye.com' and place it under directory './dataset/raw'
# Otherwise an error would pop out "Dataset not found. Please download 'DGraphFin.zip' from 'https://dgraph.xinye.com' and move it to './raw' "
from torch_geometric.datasets import DGraphFin

dataset = DGraphFin(root='./dataset')
data = dataset[0]

data
>>> Data(x=[3700550, 17], edge_index=[2, 4300999], y=[3700550], edge_type=[4300999], edge_time=[4300999], train_mask=[3700550], val_mask=[3700550], test_mask=[3700550])

Note: Please download DGraphFin dataset file 'DGraphFin.zip' on our website 'https://dgraph.xinye.com' and place it under directory './dataset/raw' before running the example, otherwise an error would pop out "Dataset not found. Please download 'DGraphFin.zip' from 'https://dgraph.xinye.com' and move it to './raw' "

Baselines

This repo provides a collection of baselines of DGraphFin. Please download the dataset file on our website and place it under the folder './dataset/DGraphFin/raw'.

Environments

Implementing environment:

• numpy = 1.21.2

• pytorch = 1.6.0

• torch_geometric = 1.7.2

• torch_scatter = 2.0.8

• torch_sparse = 0.6.9

• GPU: Tesla V100 32G

Training

To get the performance for each model, simply run the following lines of code:

• MLP

python gnn.py --model mlp --dataset DGraphFin --epochs 200 --runs 10 --device 0

• GCN

python gnn.py --model gcn --dataset DGraphFin --epochs 200 --runs 10 --device 0

• GraphSAGE

python gnn.py --model sage --dataset DGraphFin --epochs 200 --runs 10 --device 0

• GraphSAGE (NeighborSampler)

python gnn_mini_batch.py --model sage_neighsampler --dataset DGraphFin --epochs 200 --runs 10 --device 0

• GAT (NeighborSampler)

python gnn_mini_batch.py --model gat_neighsampler --dataset DGraphFin --epochs 200 --runs 10 --device 0

• GATv2 (NeighborSampler)

python gnn_mini_batch.py --model gatv2_neighsampler --dataset DGraphFin --epochs 200 --runs 10 --device 0

Performances:

Below are the performances on DGraphFin(10 runs):

Methods	Train AUC	Valid AUC	Test AUC
MLP	0.7221 ± 0.0014	0.7135 ± 0.0010	0.7192 ± 0.0009
GCN	0.7108 ± 0.0027	0.7078 ± 0.0027	0.7078 ± 0.0023
GraphSAGE	0.7682 ± 0.0014	0.7548 ± 0.0013	0.7621 ± 0.0017
GraphSAGE (NeighborSampler)	0.7845 ± 0.0013	0.7674 ± 0.0005	0.7761 ± 0.0018
GAT (NeighborSampler)	0.7396 ± 0.0018	0.7233 ± 0.0012	0.7333 ± 0.0024
GATv2 (NeighborSampler)	0.7698 ± 0.0083	0.7526 ± 0.0089	0.7624 ± 0.0081

Datasets Overview

To advance research on graph foundation models and the evolution of dynamic graphs, we continue to release real-world graph datasets. Three large-scale datasets are currently publicly available, with details provided below.

Summary

Name	Evolving Pattern Modeling	Graph Anomaly Detection	Pretraining of Dynamic Graph
DGraph-Fin	✓	✓	✓
DGraph-Fin2	✓	✓	✓
DGraph-Nearby	✓		✓

Name	Type	Time Unit	Time span	# Node	# Edge	Is Directed	Node label	Edge label	Temporal node label
DGraph-Fin	Social Network	Day	~ 2 Years	3,700,550	4,300,999	✓	✓
DGraph-Fin2	Social Network	Day	~ 2 Years	3,700,550	4,300,999	✓	✓		✓
DGraph-Nearby	Social Network	Day	~ 3 Years	21,796	190,332	✓