Overview

tseda is designed for fast exploratory analysis of regularly sampled time series (hourly or lower cadence) with Singular Spectrum Analysis (SSA).

Why SSA-first time series analysis

Before model fitting, it is often useful to separate a series into structural parts: trend or trend-cycle, seasonality, and noise. This helps you decide what signal structure is worth keeping for downstream tasks such as forecasting, anomaly detection, or monitoring.

tseda uses an SSA-first workflow because SSA gives you a high-resolution decomposition with one primary control parameter (window size). You can then inspect explained variance, component groupings, and residual behavior before choosing any downstream model.

For additional conceptual background and worked examples, see:

Core capabilities

  • Initial assessment: data profile, spread, distribution, and autocorrelation cues.

  • SSA decomposition: component extraction, grouping, reconstruction, and diagnostics, including a knee-based (commonly used elbow heuristic) noise-floor estimate for initial signal-pool selection.

  • Observation logging: AIC-by-rank summaries, editable narrative reporting, and

    validated knowledge-base save locations.

Design philosophy

The package follows a few explicit design rules:

  • UI and notebook parity: features available in the Dash UI are exposed through

    Python calls for notebook workflows.

  • Configuration over hard-coding: algorithm constants and thresholds are loaded

    from tseda_config.yaml and can be tuned without editing source.

  • Explicit decomposition controls: SSA window and component grouping are treated

    as first-class controls rather than hidden internals.

  • Composable analysis: each plotting/diagnostic feature is a separate callable,

    allowing custom notebook pipelines and automated reports.

For installation and usage details, see Installation and Workflow.