liveBook · Manning

Granularity

As explained in Machine Learning System Design

The level of detail at which data is represented, measured, or modeled.

Granularity is the level of detail chosen for representing data. In demand forecasting, that choice is also a product decision: finer records may support richer use cases, while coarser records can reduce computation without hurting performance. section 9.4.1

Daily versus weekly demand data

Representing demand by day, store, and item creates a more granular dataset. Aggregating to week, store, and item reduces required computations by a factor of seven, while performance may remain unchanged or improve. The choice should be checked with product managers, category managers, users, and other stakeholders because it reshapes the stock management system’s possible uses. section 9.4.1

How Machine Learning System Design covers it

How deep it goes
Contrasts daily and weekly aggregation, quantifies a sevenfold computation reduction, and stops short of prescribing one universal resolution.
Examples
Uses day-store-item and week-store-item demand records as the central granularity comparison.
What you can do after
Evaluate data-resolution choices against computation, validation performance, service speed, and stakeholder needs.
How it connects
Connects dataset size, learning curves, demand forecasting, model performance, and stock management systems.

Also explained in

Compare how 8 books explain granularity →

Related concepts

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.