Definition

In time-series analysis and monitoring, a window is a time interval (slice) of data that you analyze.

  • It defines how much past data you consider when computing a statistic (mean, sum, baseline, anomaly score, etc.).
  • Windows can move (rolling/sliding), stay fixed, or expand.

Types of Windows

  1. Fixed Window
    • Always refers to a fixed period, like last 7 days.
    • Example: “Compute weekly sales average from Monday–Sunday.”
  2. Rolling / Sliding Window
    • Moves forward in time step by step, recalculating metrics.
    • Example: “7-day rolling average of daily temperature.”
    • Used for smoothing, anomaly detection, baselines.
  3. Expanding Window
    • Starts at a point in time and grows as more data arrives.
    • Example: “Cumulative average sales from product launch to today.”

Example

rolling baseline (last N weeks) vs current (last 24h–7d)

  • Baseline Window (last N weeks):
    • A rolling/sliding window of N weeks, computing the historical average.
    • Used as the “normal” reference.
  • Current Window (last 24h–7d):
    • A short, recent window.
    • Used to see if current behavior deviates from the baseline.

Example (API monitoring)

  • Rolling baseline (last 8 weeks) → avg latency = 200 ms.
  • Current window (last 24h) → avg latency = 350 ms.
  • Since 350 ms >> 200 ms, this flags performance degradation.

Summary:
A window is a chosen time interval of data for calculating metrics.

  • Windows let you compare recent trends vs. historical baselines.
  • Main types: fixed, rolling, expanding.