Skip to content
ML Academy

Introduced on day 18

Error slice

A subset of evaluation rows defined by something the decision depends on (segment, route, season, lead time, volume), with its own count, bias and error.

In detail

Slices turn one aggregate metric into a map of where a model fails. Define them from information available at decision time, show count and bias beside the error, and rank them by volume x error x unit cost. Many slices invite chance findings, so pre-choose them or require intervals.