SeDyT: A General Framework for Multi-Step Event Forecasting via Sequence Modeling on Dynamic Entity Embeddings

Dependencies

• python >= 3.8.8
• pytorch >= 1.8.1
• dgl >= 0.6.1
• colorama >= 0.4.4
• numpy >= 1.19.2

Configuration Files

All configuration files are stored at /config/<DATASET>/*.yml. The meaning for each hyper-parameter is

• emb-net: embedding network
  • dim: dimension of the output embedding
  • dim_e: dimension of the learnable entity embedding
  • dim_t: dimension of the time encoding
  • history: number of history for the self connection
  • layer: number of GNN layers
  • sample: number of neighbors, '-1' for no neighbors
  • granularity: time duration for each sliding window
  • r_limit: maximum types of relationships, the other relationships are treated as an additional type
• gen-net: sequence model
  • dim_r: dimension of the learnable relationship embedding
  • arch: architecture of each layer, separated by '-'
  • dim: output dimension of each layer, separated by '-'
  • att_head: number of attention heads of each layer, separated by '-'
  • history: selected input history embeddings, separated by ' '
• train: training parameter
  • fwd: ignore this number of timestamps at beginning (for fast training)
  • epoch: number of epochs
  • batch_size: batch size
  • lr: learning rate
  • dropout: dropout rate
  • weight_decay: weight decay rate
  • norm_loss: currently deprecated

Run

To run the code

python train.py --data <DATASET> --config <path_to_config>

The first run will generate the graph, which will need more time.

Citation

Here's the bibtex in case you want to cite our work

@inbook{10.1145/3459637.3482177,
    author = {Zhou, Hongkuan and Orme-Rogers, James and Kannan, Rajgopal and Prasanna, Viktor},
    title = {SeDyT: A General Framework for Multi-Step Event Forecasting via Sequence Modeling on Dynamic Entity Embeddings},
    year = {2021},
    isbn = {9781450384469},
    publisher = {Association for Computing Machinery},
    address = {New York, NY, USA},
    url = {https://doi.org/10.1145/3459637.3482177},
    booktitle = {Proceedings of the 30th ACM International Conference on Information & Knowledge Management},
    pages = {3667–3671},
    numpages = {5}
}