TF-TGN

The codebase for TF-TGN, a temporal graph neural network (TGNN) based on the Transformer decoder to model the evolution of temporal graphs.

• TF-TGN structures the message aggregation operation between chronologically occurring nodes and their temporal neighbors in TGNNs as sequence modeling.
• TF-TGN unified the traning paradigms of TGNNs as as sequence modeling and fill the gap between TGNNs (or even GNNs) and LLMs.

Sampling

This code is the implementation of the parallel sampling algorithm in the paper to enhance the efficiency of T-CSR converting and temporal neighbor sampling.

compile

g++ -fopenmp -std=c++11 main.cpp TemporalGraph.cpp readcsv.cpp  utils.cpp -o main

string file_path in the main.cpp is the path of the dataset.

For example, string file_path = "./reddit.csv", reddit.csv is the Reddit dataset.

run

For example, sample_num=128, batch_size=512

./main 128 512

TF-TGN

The code of the TF-TGN model will be released once we obtain open-source licenses from our collaborators and partner organizations.

TF-TGN incorporates the the following components:

• Flash-attention
• Memeory-efficient attention
• PyTorch FSDP
• Mixed precision training