This is the code for HATS: A Hierarchical Sequence-Attention Framework for Inductive Set-of-Sets Embeddings

We focus on Set-of-Sets problems and evaluate our model in arithmatic set operation task, Adamic/Adar Index, subgraph hyperlink prediction and set of pointclouds classification.

The first Set-of-Sets task is to perform arithmetic on sequences of integers: intersection sum, intersection binary, unition sum.

The second Set-of-Sets task learns to compute Adamic-Adar Index.

The third Set-of-Sets task predicts whether there is a hyperlink between two set of nodes. The feature of each node is the embedding vectors from [GraphSAGE].

The fourth Set-of-Sets task predicts among a set of pointclouds a) whether there is a pointcloud with different label. b) the number of unique labels.
Please see the supplementary section for a brief description and summary of the code.

• PyTorch 0.4.0 or later - which can be downloaded here
• Python 3.7

For the sequence based tasks, please use the following format:

• python train.py -m "model name" -t "task" -l "number of hidden layers in rho" -lr "learning rate" -b "batch size" -i "running times"
• Permitted models are {deepsets,lstm,gru,cnn,hats,hier}

For the graph based tasks, we have provided an example below:

• python train.py -m hats -t inter_sum -i 2 -lr 0.001

We recommend training these models on a GPU.

• For the arithmetic tasks, the data is generated on the fly as described in our paper. You can adjust the number of training, test and validation examples used.
• For the Adamic/Adar and hyperlink prediction tasks, the following datasets were used:
  • PPI
  • Cora, available at the GraphSAGE PyTorch repo
  • Pubmed, available at the GraphSAGE PyTorch repo
• For pointcloud prediction tasks, we used ModelNet40 dataset (http://modelnet.cs.princeton.edu/)

Please feel free to reach out to Changping Meng (meng40 at purdue.edu) if you have any questions.

If you use this code, please consider citing:

@inproceedings{meng2019hats,
  title={HATS: A Hierarchical Sequence-Attention Framework for Inductive Set-of-Sets Embeddings},
  author={Meng, Changping and Yang, Jiasen and Ribeiro, Bruno and Neville, Jennifer},
  year={2019},
  organization={KDD}
}