Skip to content

Latest commit

 

History

History
76 lines (54 loc) · 4.21 KB

README.md

File metadata and controls

76 lines (54 loc) · 4.21 KB
Raw

TAP-GNN

Code for Temporal Aggregation and Propagation Graph Neural Networks For Dynamic Representation, and we have updated the latest results of baseline methods in the arxiv version than the IEEE published version.

Authors: Tongya Zheng, Xinchaon Wang, Zunlei Feng, Jie Song, Yunzhi Hao, Mingli Song, Xingen Wang, Xinyu Wang, Chun Chen

We utilize the cool feature group_apply_edges of dgl-0.4.3post2 to compute the aggregated neighbor messages of nodes along the chronological order, reducing the computation complexify of temporal graph convolution from $O(\sum_{v \in V} d_v^2) \ge O(E\cdot \bar{d})$ to $O(E)$. The proposed TAP-GNN reuses the historical node embeddings to compute new node embeddings, eliminating nodes' cross-time dependencies on historical neighbors.

The latest dgl version has been dgl-1.0x by now and removed the slow operation group_apply_edges since 0.4.3. We are trying to work out our TAP-GNN with the newly implemented segment_reduce feature.

Setup the Environment

  • conda create -n tap python=3.7 -y

  • pip install -r requirements.txt

  • My torch version is torch-1.9.1+cu102

Data Preprocess

We have preprocessed most temporal graphs in the data/format_data directory, and placed the JODIE datasets at Google drive, which can be downloaded and placed at the data/format_data.

bash init.sh

We use init.sh to make necessary directories for our experiments to store generated datasets by data/*, boost the training speed by gumbel_cache and sample_cache, record training details by log, record testing results by results and nc-results, save our trained models by ckpt and saved_models.

python data_unify.py -t datasplit
python data_unify.py -t datalabel

We use -t datasplit to split datasets into the training, validation and testing set according to the ratios.

Run the Script

Running bash ./init.sh to create necessary directories.

Temporal Link Prediction (Transductive)

  • We have computed the target edge ids of the proposed Aggregation and Propagation (AP), where we store the temporal node embeddings with the edge-size hidden embeddings. It is the same with our analysis in the paper that the size of temporal node embeddings is at the same-level of temporal edges $O(E)$. In this way, the Aggregation operation acts as a message-passing operation, and the Propagation operation acts as a reduction function with the proper normalization term, which can be seen in

python tap_gnn.py -d fb-forum

Online Link Prediction

  • We have built an online model with the same functionality to the full-graph training model.

python online_tap_gnn.py -d JODIE-wikipedia -bs 500 --n-layers 30

Temporal Node Classification

  • We find that using trainable node embeddings for inductive learning harms the generalization ability. So we freeze the node embeddings as all zero vectors for JODIE datasets.

python tap_gnn.py -d JODIE-wikipedia --task node --no-trainable

  • We have set the balanced training strategy for the extremely imbalanced task of temporal node classification.

python node_tap_gnn.py -d JODIE-wikipedia

Cite us

@article{zheng2023temporal,
  title={Temporal Aggregation and Propagation Graph Neural Networks For Dynamic Representation},
  author={Zheng, Tongya and Wang, Xinchao and Feng, Zunlei and Song, Jie and Hao, Yunzhi and Song, Mingli and Wang, Xingen and Wang, Xinyu and Chen, Chun},
  journal={IEEE Transactions on Knowledge and Data Engineering},
  year={2023},
  pages={1--15},
  publisher={IEEE}
}