Is the selection itself formed based on some time slices or taken for the entire period?

Data sampling can be formed in different ways depending on the task and data source:

• By time slices — when data is taken for a specific period (for example, the last month, week, or day). This is convenient for trend analysis, seasonality, or changes over time.

• All the time — when the sample includes all available data from the beginning of their accumulation. This approach is useful for general analysis or building models on the full volume of information.

In machine learning practice, time slices are often used to avoid data leakage from the future into training and to ensure correct model evaluation. However, the specific approach depends on the task and data requirements.