This is a PyTorch and TensorLy implementation of Accurate Online Tensor Factorization for Temporal Tensor Streams with Missing Values (CIKM 2021).
- Install all of the prerequisites
- You can run the demo script by
bash demo.sh
, which simply runssrc/main.py
. - You can change the dataset by modifying
src/main.py
and check the dataset indata
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.
- 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 |
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}
}