Validation window

As explained in Time Series Forecasting Using Foundation Models

A validation window is a contiguous held-out time period on which a forecast is generated during cross-validation.

A validation window is a contiguous held-out time period used to generate forecasts during time-series cross-validation. Each window tests a fixed forecast horizon against subsequently observed actual values. section 3.6

Use successive windows to approximate repeated forecasting in practice:

• Choose a forecast horizon and number of windows. section 3.6
• Generate forecasts, then update the input with the actual values from that window. section 3.6
• Compare predictions with actuals across windows using metrics such as MAE and sMAPE. section 6.3.2

Figure 3.14 Illustrating the behavior of cross-validation with TimeGPT. A fold designates one round of cross-validation. By default, nonoverlapping windows are used.

A diagram shows each fold as a fixed-horizon forecast and distinguishes the successive nonoverlapping validation windows. section 3.6

How Time Series Forecasting Using Foundation Models covers it

How deep it goes

Explains sequential updates, fixed horizons, nonoverlap, fold counts, and metric evaluation, but does not develop the underlying forecasting models.

Examples

Uses Walmart weekly sales with four eight-week windows, producing 32 test points for model comparison.

What you can do after

Evaluate forecast performance across repeated test periods using MAE and sMAPE.

How it connects

Connects cross-validation with forecast horizon, folds, MAE, sMAPE, and model evaluation.

liveBookmanning.com

Read Time Series Forecasting Using Foundation Models on liveBook Get the book at manning.com

Related concepts

• cross-validationpart of · Cross-validation is a resampling technique that evaluates a model by repeatedly fitting it on some portions of available data and assessing it on held-out portions.
• forecast horizonconfigured by · A forecast horizon is the future time span or number of time steps over which predictions or outcomes are generated or evaluated.
• time windowcontrasts with · A bounded temporal interval or event scope over which observations or events are grouped, evaluated, or processed.

This explainer was generated by Manning's knowledge graph from the text of Time Series Forecasting Using Foundation Models. The section links point to the parts of the book it draws on.