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: https://doi.org/10.1007/978-3-031-70344-7_25

• 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.

@InProceedings{10.1007/978-3-031-70344-7_25,
	author="Phungtua-eng, Thanapol and Yamamoto, Yoshitaka",
	editor="Bifet, Albert and Davis, Jesse and Krilavi{\v{c}}ius, Tomas and Kull, Meelis and Ntoutsi, Eirini and {\v{Z}}liobait{\.{e}}, Indr{\.{e}}",
	title="Adaptive Seasonal-Trend Decomposition for Streaming Time Series Data with Transitions and Fluctuations in Seasonality",
	booktitle="Machine Learning and Knowledge Discovery in Databases. Research Track",
	year="2024",
	publisher="Springer Nature Switzerland",
	address="Cham",
	pages="426--443",
	isbn="978-3-031-70344-7"
}

Link

• Our laboraory at Shizuoka University

Contact

If you have any question, please contact thanapol@yy-lab.info

Acknowledgements

This work was supported by JSPS KAKENHI Grant Numbers JP20K11935, JP21H04491 and 24K15086. We would also like to thank the anonymous reviewers for giving us useful and constructive comments. Additionally, we are grateful to the community and everyone who made their datasets and source codes publicly accessible. These datasets and source codes are valuable and have greatly facilitated this research.