ExplanationLoadScenario
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.
Reasons owed for every denial: decisions_per_day at denial_share, each explained by a sampling attributor of samples model evaluations or by exact Shapley over features.
Notes
Equation: Sampling work is denials*samples; exhaustive subset work is denials*2**features.
Assumptions: Every denial gets one explanation; exact work counts one model evaluation per feature subset.
Excludes: Memoization, specialized Shapley algorithms, explanation quality, and non-model processing cost.
References
Derivation: Sampling work is denials*samples; exhaustive subset work is denials*2**features. This is declared scenario accounting rather than a fitted empirical law.
Examples
>>> from mlsysim import Scenarios
>>> scenario = Scenarios.CreditDenialExplanationLoad
>>> value = scenario.notices_per_day
>>> round(float(value), 3)
475000.0Attributes
| Name | Description |
|---|---|
| decisions_per_day | |
| denial_share | |
| exact_evaluations | Daily exhaustive coalition-evaluation count across explained denials. |
| exact_per_explanation | Coalition count 2**features for exhaustive subset evaluation, without implementation |
| features | |
| notices_per_day | Expected daily denials requiring an explanation. |
| sampled_evaluations | Daily model-evaluation count for fixed samples per explained denial. |
| sampled_over_decisions | Attribution model evaluations per original decision, dimensionless. |
| samples |
decisions_per_day
Type: float.
Declared default: required.
samples
Type: int.
Declared default: required.
features
Type: int.
Declared default: required.
notices_per_day
engine.scenario_types.ExplanationLoadScenario.notices_per_day
Expected daily denials requiring an explanation.
Equation: decisions_per_day * denial_share.
sampled_evaluations
engine.scenario_types.ExplanationLoadScenario.sampled_evaluations
Daily model-evaluation count for fixed samples per explained denial.
Equation: notices_per_day * samples.
sampled_over_decisions
engine.scenario_types.ExplanationLoadScenario.sampled_over_decisions
Attribution model evaluations per original decision, dimensionless.
Equation: sampled_evaluations / decisions_per_day.
exact_per_explanation
engine.scenario_types.ExplanationLoadScenario.exact_per_explanation
Coalition count 2**features for exhaustive subset evaluation, without implementation reuse.
Equation: 2 ** features.
exact_evaluations
engine.scenario_types.ExplanationLoadScenario.exact_evaluations
Daily exhaustive coalition-evaluation count across explained denials.
Equation: notices_per_day * exact_per_explanation.