OutputShapingScenario
Reliability scenarios · Implementation: mlsysim/engine/scenario_definitions/reliability.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
An API that returns the top top_k of class_count scores, rounded from from_decimals to to_decimals decimal places.
Notes
Equation: Entry fractions count retained scores; log-two level counts express decimal representation sizes in bits.
Assumptions: Positive top-k count and decimal rounding of scores in [0,1].
Excludes: An attack-success model or information-theoretic privacy guarantee.
References
Derivation: Entry fractions count retained scores; log-two level counts express decimal representation sizes in bits. 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.ImageNetTop5RoundedScores
>>> round(float(scenario.scores_reduction), 4)
200.0Attributes
| Name | Description |
|---|---|
| bits_removed | Log-two count of the removed decimal bins, in bits; this is a representation count, not |
| class_count | |
| decimal_digits_removed | Number of decimal places removed from each returned score. |
| entries_removed | Fraction of score entries suppressed by returning only the top-k entries. |
| from_decimals | |
| retained_bits | Log-two count of the retained score levels including both endpoints of [0,1]. |
| scores_reduction | Dimensionless full-score count divided by returned top-k score count. |
| to_decimals | |
| top_k |
class_count
Type: int.
Declared default: required.
top_k
Type: int.
Declared default: required.
from_decimals
Type: int.
Declared default: required.
to_decimals
Type: int.
Declared default: required.
scores_reduction
engine.scenario_types.OutputShapingScenario.scores_reduction
Dimensionless full-score count divided by returned top-k score count.
Times fewer scores per answer than returning every class.
Equation: class_count / top_k.
entries_removed
engine.scenario_types.OutputShapingScenario.entries_removed
Fraction of score entries suppressed by returning only the top-k entries.
Equation: 1 - top_k / class_count.
decimal_digits_removed
engine.scenario_types.OutputShapingScenario.decimal_digits_removed
Number of decimal places removed from each returned score.
Equation: from_decimals - to_decimals.
bits_removed
engine.scenario_types.OutputShapingScenario.bits_removed
Log-two count of the removed decimal bins, in bits; this is a representation count, not a privacy guarantee.
Bits per score removed under a decimal-bin count.
Equation: calc_level_bits(10 ** decimal_digits_removed).
retained_bits
engine.scenario_types.OutputShapingScenario.retained_bits
Log-two count of the retained score levels including both endpoints of [0,1].
Bits left in a score in [0, 1] at to_decimals places (10^d + 1 levels).
Equation: calc_level_bits(10 ** to_decimals + 1).