Adaptive Seasonal-Trend Decomposition (ASTD)

• This is an official implementation of "Adaptive Seasonal-Trend Decomposition for Streaming Time Series Data with Transitions and Fluctuations in Seasonality" (ASTD).

• ** This paper was accepted in ECML-PKDD 2024 **

• ASTD was implemented using Python 3.9.2.

• Existing methods were implemented using Python 3.9.2 and R 4.3.0

• If you have any more questions or need further suggestions, don't hesitate to email me.

Document material

• Supplementary website

Folder structure

.
├── datasets                            # Datasets were utilized in this paper. 
│   ├── 01_synthetic_datasets           
│   ├── 02_Real1_datasets             
│   ├── 03_Real2_datasets
│   └── README.md                      # Readme for dataset folder
├── document                           # Supplmentary file
├── evaluation                         # All source codes for reproduction
│   ├── 00_HAQSE                       # Reproduction for HAQSE estimator
│   ├── 01_synthetic_datasets          # Reproduction for synthetic datasets
│   ├── 02_Real1_datasets              # Reproduction for Real1
│   └── utility_evaluation.R           # Utility function for evaluation with R
├── figures                            # Reproduction for Figures in this paper 
├── src                                # Source files
│   ├── utilities                      # Utility functions
│   └── online_decomposition           # Online Time series decomposition
└── README.md

Note that our datasets were cleaned in the same format, but we give information to access the original sources.

Dependencies

Python

• [numpy] >=1.26.4
• [scipy] >= 1.12.0
• [pandas] >= 2.2.1
• [matplotlib] >= 3.8.2
• tqdm >= 4.64.1
• statsmodel >= 0.14.1
• periodicity-detection >= 0.1.1 ** For existing SLE methods
• scikit-learn >= 1.4.1 ** For mean square error computation in STD evaluations.
• rpy2 >= 3.5.16 ** For R running in python

R

• sazedR >= 2.0.2 ** SAZED method
• astsa >= 2.1 ** CRAN dataset
• fpp2 >= 2.5 ** CRAN dataset
• expsmooth >= 2.3 ** CRAN dataset
• fma >= 2.5 ** CRAN dataset

** Please be careful some libraries overwrite some datasets.

BibTex

• If you plan to use or apply our source code, please cite our published paper. Note that the DOI and BibTeX will be updated after our publication appears online.

@inproceedings{ASTD_ECMLPKDD,
	author = {Phungtua-Eng, T. and Yamamoto, Y.},
	booktitle = {Machine Learning and Knowledge Discovery in Databases: Research Track (ECML PKDD 2024)},
	title = {Adaptive Seasonal-Trend Decomposition for Streaming Time Series Data with Transitions and Fluctuations in Seasonality},
	year = {2024},
	note = {(To appear)}
}

Link

• Our laboraory at Shizuoka University

Contact

If you have any question, please contact [email protected]

Acknowledgements

We would like to thank the community and everyone who made their datasets and source codes publicly accessible. These datasets are valuable and have greatly facilitated this research.