LANTERN-NeurIPS-2019

Source code for NeurIPS 2019 paper "Learning Latent Processes from High-Dimensional Event Sequences via Efficient Sampling"

Environment

• Python 3.5
• PyTorch 1.0.1

Requirements

• GPUs with 12GB memory

Datasets

• Use data_generate.py to generate the synthetic datasets
• The memetracker dataset can be downloaded from：https://snap.stanford.edu/data/memetracker9.html
• The weibo dataset can be downloaded from: https://www.aminer.cn/influencelocality
• Our great thanks to authors of the real-world datasets.
• Use data/#DATA#/preprocess.py to preprocess the downloaded dataset and you can get the .pkl files in each folder
• To check the statistics of each dataset, run cck.py in each folder, where you could also add lines of your own code to check the dataset.

Quick Start

To train on small datasets (Syn-Small and Memetracker), you can run

python train_small.py

To train on large datasets (Syn-Large and Weibo), you can run

python train_large.py

We also released our pre-trained model parameters for each dataset in /model folder. For a quick test, run

python test.py

Citation

If you have any problems on this code, feel free to contact [email protected]. If you use this code as part of your research, please cite the following paper:

@inproceedings{LANTERN-19,
  author    = {Qitian Wu and Zixuan Zhang and Xiaofeng Gao and Junchi Yan and
               Guihai Chen},
  title     = {Learning Latent Process from High-Dimensional Event Sequences via Efficient Sampling},
  booktitle = {Thirty-third Conference on Neural Information Processing Systems, {NeurIPS} 2019, Vancouver, Canada,
               Dec 8-14, 2019},
  year      = {2019}
  }