Skip to content

Zh-Shuai/NJDTPP

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Neural Jump-Diffusion Temporal Point Processes

The implementation of our ICML-2024 (Spotlight) paper "Neural Jump-Diffusion Temporal Point Processes".

Updates

Updates (July 18, 2025)

1. Improved Training Speed: The training process now performs computations only on valid positions within padded sequences, eliminating unnecessary calculations for padding tokens. This optimization effectively reduces computational overhead, especially for datasets with large variations in sequence length.

2. Improved Inference Speed: Inference has been optimized to handle batch predictions at once, rather than processing one sequence at a time, resulting in significantly faster inference times.

3. Upper Limit Estimation for Integral: The upper limit ($\infty$) of the integral in Eq.(21) of our paper has been truncated using training data instead of test data. This change ensures a more reasonable approach.

Dataset

The real-world datasets are from "EasyTPP" and "NHP".

Installation

  1. Install the dependencies
conda env create -f environment.yml
  1. Activate the conda environment
conda activate NJDTPP
  1. Unzip the data
unzip data.zip

Reproducing the results from the paper

Go to the source directory:

cd experiments

This directory contains all experiments on three synthetic and six real-world datasets, for example:

  • MIMIC-II dataset
python mimic2.py

Citation

If you find this code useful, please consider citing our paper:

@inproceedings{zhang2024neural,
  title={Neural Jump-Diffusion Temporal Point Processes},
  author={Zhang, Shuai and Zhou, Chuan and Liu, Yang and Zhang, Peng and Lin, Xixun and Ma, Zhi-Ming},
  booktitle={International Conference on Machine Learning},
  year={2024}
}

Acknowledgements and References

Parts of this code are based on and/or copied from the code of "NJSDE" and "SAHP".

About

Implementation of "Neural Jump-Diffusion Temporal Point Processes" (ICML 2024 Spotlight)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages