Why De-Time
De-Time exists because decomposition workflows are often split across single-method libraries, notebook snippets, and benchmark-only artifacts. That fragmentation makes it harder to compare methods, preserve metadata, and move from exploratory work to something another researcher can rerun.
De-Time responds to that problem as software, not as a new method. It standardizes:
- how decomposition is configured,
- how results are returned,
- how CLI and Python usage align,
- how native acceleration and optional backends are surfaced.
The project is deliberately narrow. It aims to be a good decomposition package, not a general forecasting framework and not a benchmark leaderboard.
That narrow scope is still relevant to machine learning workflows. In practice, decomposition is often used before downstream modeling for denoising, feature extraction, representation shaping, and shared-structure analysis across channels. De-Time focuses on making those workflow steps reproducible rather than on claiming a new decomposition algorithm.