RolloutExposureScenario
Economics scenarios · Implementation: mlsysim/engine/scenario_definitions/economics.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
Users exposed until rollback under an all-at-once cutover against a canary share, with the same request rate, harm, and time to rollback.
Notes
Equation: Exposure scales with traffic share, so equal-window exposure ratio is full_share/canary_share.
Assumptions: Positive canary share, identical request rate, harm per request, and rollback duration.
Excludes: Detection-time differences between rollout policies and heterogeneous user harm.
References
Derivation: With identical rate, harm per exposed request, and rollback time, both exposure totals contain the same factors. Canceling them leaves the full-traffic share divided by the canary share.
Examples
>>> from mlsysim import Scenarios
>>> s = Scenarios.CanaryVersusCutoverExposure
>>> s.exposure_ratio > 0
TrueAttributes
| Name | Description |
|---|---|
| canary_share | |
| exposure_ratio | Dimensionless cutover-to-canary traffic exposure ratio for identical rollback windows. |
| full_share | |
| time_to_rollback |
time_to_rollback
Type: Optional[Quantity].
Declared default: None.
exposure_ratio
engine.scenario_types.RolloutExposureScenario.exposure_ratio
Dimensionless cutover-to-canary traffic exposure ratio for identical rollback windows.
Equation: full_share / canary_share.