A sequence is data ordered by time. Neighboring observations, such as daily sales or recent user actions, can share information, so future data must not inform predictions about the past. section 7.2.4
When working with time-series data, consider two practical uses:
- Validate models with splits ordered by timestamp, then choose training and test windows that reflect forecast horizons, retraining schedules, and seasonality. section 7.2.4
- Create features from past values using lags, time-window aggregates, or statistics such as standard deviation and variance. section 11.1.2
Figure 7.4 Standard time-based split. The test dataset always follows the train one, so train samples are “past” and test is “future.”
A diagram of the basic rule: training data comes from the past, and testing data follows it in the future. section 7.2.4
