Decomposition Research and Practical Guides

Use these guides to choose a method, understand its assumptions, and design a comparison that answers your application question. The articles are written by the DeTime maintainers and link to primary papers and official software sources.

Start with your question Guide
Which method and Python package fit my data? STL, SSA, Wavelets, EMD, or VMD? Choosing Python Decomposition Tools
Where is decomposition used in recent research? Time-Series Decomposition in 2026: Algorithms, Tokens, and Losses
How can components become forecasting features? History-only features and held-out evaluation
Which methods does DeTime actually implement? Method matrix and configuration reference

Download the reading list

The initial selection contains 20 references, with publication records checked on 29 October 2026. Eight cover classical methods and reference material, eight cover neural architectures, tokenization, objectives and decoders, and four cover methods or evaluation with 2026 publication records.

The map identifies literature candidates separately from implemented DeTime features. It records publication status and the cited version rather than assuming that a preprint date is the final publication date. The selection is intended to support method choice and follow-up reading; it is not exhaustive.

Use and share

The quickstart provides the first runnable decomposition. The English sharing kit contains a community post, short post, thread, and newsletter blurb linked to the comparison guide. The drafts disclose maintainer involvement and are ready to adapt to the destination's format and community rules.

Suggest an addition

Open a repository issue with the publication title, primary-source URL, publication status, and a brief explanation of the decomposition mechanism. For software comparisons, include the upstream documentation and tested version when available. An implementation or performance claim needs its own reproducible evidence.