STF

This is a PyTorch and TensorLy implementation of Accurate Online Tensor Factorization for Temporal Tensor Streams with Missing Values (CIKM 2021).

Prerequisites

• Python 3.6+
• DotMap
• NumPy
• pandas
• scikit-learn
• PyTorch
• TensorLy

Usage

• Install all of the prerequisites
• You can run the demo script by bash demo.sh, which simply runs src/main.py.
• You can change the dataset by modifying src/main.py and check the dataset in data directory.
• You can change the number of hyper-parameters by modifying src/stf.py.
• you can check out the running results in out directory, and then plot the results.

Datasets

• There are six data files stored in COO format (e.g. i j k value), and data statistics as follows.

Name	Description	Size	NNZ	Granularity in Time	Original Source
Beijing Air Quality	locations x pollutants x time	12 x 6 x 5994	618835	hourly	Link
Madrid Air Quality	locations x pollutants x time	26 x 17 x 3043	383279	hourly	Link
Radar Traffic	locations x directions x time	17 x 5 x 6419	181719	hourly	Link
Indoor Condition	locations x sensor x time	9 x 2 x 2622	59220	hourly	Link
Intel Lab Sensor	locations x sensor x time	54 x 4 x 1152	513508	every 10 minutes	Link
Chicago Taxi	sources x destinations x time	77 x 77 x 2904	424440	hourly	Link

Reference

If you use this code, please cite the following paper.

@inproceedings{ahn2021accurate,
  title={Accurate Online Tensor Factorization for Temporal Tensor Streams with Missing Values},
  author={Ahn, Dawon and Kim, Seyun and Kang, U},
  booktitle={Proceedings of the 30th ACM International Conference on Information \& Knowledge Management},
  pages={2822--2826},
  year={2021}
}