A time window sets how many observations a time-series training or testing segment contains. For testing, choose a size that matches the forecast distance and expected time in production before retraining. Window size is also a model-complexity choice: compare candidate granularities with a model-wise learning curve. section 7.2.4 section 9.4.1
Time window
The number of observations included in a time-series training or testing window.
How Machine Learning System Design covers it
- How deep it goes
- Relates test-window sizing to forecast and retraining needs, and tunes candidate granularities with a model-wise learning curve.
- Examples
- Compares window sequences such as 7/14/21 and 30/60/90, and relates test size to forecast distance.
- What you can do after
- Evaluate candidate window granularities with a model-wise learning curve and align test spans with forecast and retraining needs.
- How it connects
- Links time-series validation to forecasting, retraining, seasonality, and model complexity.
Also explained in
-
Kafka Streams in Action, Second EditionWindow is a Kafka Streams API type representing the time interval associated with a windowed record or aggregation.
-
Logs and TelemetryAn event window is a defined scope of multiple stream events over which a query computes filtering or aggregate results.
-
Apache Kafka in ActionA time window is a bounded temporal interval over which stream records are aggregated, correlated, or otherwise processed.
-
MongoDB 8.0 in Action, Third EditionA time window partitions streaming data into a bounded set of records delimited by time for stateful processing.
-
Kafka for ArchitectsA time window is a bounded interval of event or processing time over which a stream processor groups and evaluates records.
Compare how 6 books explain time window →
Related concepts
- hopping windowincludes · A hopping window is a fixed-duration time window that advances at regular intervals and may overlap with other windows.
- tumbling windowincludes · A tumbling window is a fixed-duration, non-overlapping stream-processing interval in which each event belongs to exactly one window.
- sliding windowincludes · A sliding window is a bounded region over an ordered sequence or time range that advances, updating which elements fall within it.
- TimeWindowsproduced by · TimeWindows is a Kafka Streams class that defines fixed-size time-based windows with a configurable advance interval.
- training datasetconfigures · A collection of examples or observations supplied to a learning procedure to fit or adapt a model or learn its patterns or parameters.
This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.