Module monitor 

Monitoring a stored model on new periods: actual against expected.

As a model’s predictions meet actuals, period by period, the questions are whether any period is off by more than noise and whether the ratio is drifting. actual_vs_expected answers both from the model’s own variance function: each period’s A - E is scaled by its standard deviation under the model, and a weighted trend of A / E - 1 across periods tests for drift. lift does the same split by predicted rate instead of by period.

Structs

Monitor

The result of actual_vs_expected.

PeriodSummary

One period’s (or the total’s) actual against expected.

Functions

actual_vs_expected

Actual against expected by period, with each period’s z-score under the model (family’s variance function and dispersion) and a test for drift.