liveBook · Manning

Count data

As explained in Machine Learning System Design

Data consisting of nonnegative integer-valued observations, often recorded over time as counts of events or units.

Count data records events or units as integer-valued observations. In retail, each item-store-period demand forecast is count data: the system predicts how many units will sell while keeping delivered and sold quantities close and avoiding stockouts. section 5.3.1

Treat count data as a forecasting-design constraint:

  • Use discrete predictive distributions when decisions depend on the full range of possible demand, not only a point forecast. section 5.3.1
  • Question continuous-distribution approximations when counts become small or granular. section 5.3.1
  • Judge metrics against the operational goal: narrowing the delivered-versus-sold gap while meeting a defined service-level agreement. section 5.3.1

How Machine Learning System Design covers it

How deep it goes
Defines integer-valued time series and motivates predictive distributions, but stops short of deriving specific count models.
Examples
Uses item-store-period demand, sold units, delivered units, stockouts, and service-level agreements as examples.
What you can do after
Evaluate count forecasts using metrics aligned with discrete demand distributions and inventory-service goals.
How it connects
Connects count data with time series, demand forecasting, predictive distributions, supply chain forecasting, and inventory control.
Cover of Machine Learning System Design
liveBookmanning.com
Read Machine Learning System Design on liveBook Get the book at manning.com

Also explained in

Compare how 2 books explain count data →

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.