This authors' official PyTorch implementation for paper: Streaming Factor Trajectory Learning for Temporal Tensor Decomposition(NeurIPS 2023)

To model the temporal tensor data, we assign each tensor factor with Temporal-Gaussian Process priors and model it as the continuous-time-varing trajectory.

The inference of the factor trajectory is in a streaming and online manner.

Example of learned functional frajectories of factors from real-world data.

The project is mainly built with pytorch 1.10.1 under python 3. Besides that, make sure to install tqdm and tensorly before running the project.

1. Clone this repository.
2. To play with the code quickly with visulization of factors, we offer several notebooks at notebook(on synthetic & real data)
3. To run the real-world datasets with scripts, see my_script.sh for example.
4. To tune the (hyper)parametrs of model, modify the .yaml files in config folder
5. To apply the model on your own dataset, please follow the ./data/process_script/beijing_15k_process.ipynb or ./data/synthetic/simu_data_generate_CP_r1.ipynb to process the raw data into appropriate format.

Please check our paper for more details.

Please cite our work if you would like it

@misc{fang2023streaming,
      title={Streaming Factor Trajectory Learning for Temporal Tensor Decomposition}, 
      author={Shikai Fang and Xin Yu and Shibo Li and Zheng Wang and Robert Kirby and Shandian Zhe},
      year={2023},
      eprint={2310.17021},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}