DpTrainingPlanScenario
Training scenarios · Implementation: mlsysim/engine/scenario_definitions/training.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
An illustrative differential-privacy planning scenario for one training run of base_cost.
Each privacy budget is an assumed accountant result; the stronger budget is assumed to need cost_multiplier times the training, with the stated accuracy drops and overheads (percent) as planning assumptions.
Notes
Equation: Apply the declared multiplier to baseline cost; it is not derived from the target privacy budget.
Assumptions: The privacy cost multiplier is a supplied illustrative planning input.
Excludes: A DP accountant, epsilon certification, and measured training slowdown.
References
Derivation: Apply the declared multiplier to baseline cost; it is not derived from the target privacy budget. Individual property contracts identify the exact accounting expression or reusable calculation. Registry entries supply the scenario inputs; this class does not turn them into measured outcomes.
Examples
Evaluate a named scenario using its declared inputs.
>>> from mlsysim import Scenarios
>>> scenario = Scenarios.Gpt3PrivacyPlan
>>> round(float(scenario.cost_strong.magnitude), 4)
13800000.0Attributes
| Name | Description |
|---|---|
| accuracy_drop_moderate | |
| accuracy_drop_strong | |
| base_cost | |
| clipped_share | |
| cost_multiplier | |
| cost_strong | USD training cost under the supplied privacy cost multiplier; no privacy accountant is |
| epsilon_moderate | |
| epsilon_strong | |
| epsilon_weak | |
| overhead_moderate |
epsilon_strong
Type: float.
Declared default: required.
epsilon_moderate
Type: float.
Declared default: required.
epsilon_weak
Type: float.
Declared default: required.
base_cost
Type: Quantity.
Declared default: required.
cost_multiplier
Type: float.
Declared default: required.
accuracy_drop_strong
Type: float.
Declared default: required.
accuracy_drop_moderate
Type: float.
Declared default: required.
overhead_moderate
Type: float.
Declared default: required.
cost_strong
engine.scenario_types.DpTrainingPlanScenario.cost_strong
USD training cost under the supplied privacy cost multiplier; no privacy accountant is evaluated.
Equation: (base_cost * cost_multiplier).to('USD').