Pytorch implementation for Prompt-augmented Temporal Point Process for Streaming Event Sequence, NeurIPS 2023.

First, please make sure you have an environment compatible with the following requirement

torch == 1.11.0
numpy
pandas

Lower version of pytorch should also be working but we have not tested it.

Assume we are running PT-attNHP over the Amazon data and setup the config files.

Step 1: We need to configure the parameter file corresponding to the dataset

vim dataset_config.yaml

NOTE: in example_config/dataset_config.yaml, one needs to setup information of the dataset, where we have put the default params of Amazon there.

Step 2: we need to choose the TPP model and configure the parameter file corresponding to the model

vim model_config.yaml

NOTE: in example_config/model_config.yaml, one needs to setup information of the model specs, where we have put the default params of PT-attNHP there.

Step 3: Then we train the chosen TPP model and evaluate

python run_pt_anhp.py

If you find this repository useful for your work, please consider citing it as follows:

@inproceedings{xue2023prompt,
  title={Prompt-augmented Temporal Point Process for Streaming Event Sequence},
  author={Xue, Siqiao and Wang, Yan and Chu, Zhixuan and Shi, Xiaoming and Jiang, Caigao and Hao, Hongyan and Jiang, Gangwei and Feng, Xiaoyun and Zhang, James Y and Zhou, Jun},
  booktitle={Advances in Neural Information Processing Systems},
  year={2023},
  url={https://arxiv.org/abs/2310.04993}
}

The following repositories are used in our code, either in close to original form or as an inspiration:

• EasyTemporalPointProcess
• Neural Hawkes Process
• HYPRO TPP