Class UncertaintyQuantificationOptions

Namespace

AiDotNet.Models.Options

Assembly

AiDotNet.dll

Configuration options for enabling uncertainty quantification during inference.

public sealed class UncertaintyQuantificationOptions

Inheritance

object

UncertaintyQuantificationOptions

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

Uncertainty quantification (UQ) augments standard point predictions with an uncertainty estimate. For supported model types, the library can sample multiple stochastic predictions and aggregate them into a mean prediction and an uncertainty estimate (variance).

For Beginners: This lets you ask the model not only "what is the prediction?" but also "how sure are you?"

Properties

AdaptiveConformalBins

Gets or sets the number of confidence bins used when ConformalMode is Adaptive.

public int AdaptiveConformalBins { get; set; }

Property Value

int

CalibrationMethod

Gets or sets the probability calibration method used when calibration labels are provided.

public ProbabilityCalibrationMethod CalibrationMethod { get; set; }

Property Value

ProbabilityCalibrationMethod

ConformalConfidenceLevel

Gets or sets the desired conformal coverage level when using conformal prediction.

public double ConformalConfidenceLevel { get; set; }

Property Value

double

Remarks

This is typically expressed as a probability (e.g., 0.9 for 90% coverage).

ConformalMode

Gets or sets the conformal calibration mode used when producing conformal intervals/sets.

public ConformalPredictionMode ConformalMode { get; set; }

Property Value

ConformalPredictionMode

Remarks

Split conformal is the standard baseline. Cross-conformal can improve stability on small calibration sets. Adaptive conformal adjusts thresholds based on predicted confidence buckets.

CrossConformalFolds

Gets or sets the number of folds used when ConformalMode is CrossConformal.

public int CrossConformalFolds { get; set; }

Property Value

int

DeepEnsembleInitialNoiseStdDev

Gets or sets the standard deviation of the initial parameter perturbation applied when constructing ensemble members.

public double DeepEnsembleInitialNoiseStdDev { get; set; }

Property Value

double

Remarks

This helps ensure ensemble members do not collapse to identical solutions when created from a shared base model.

DeepEnsembleSize

Gets or sets the number of independently trained models used for deep ensemble uncertainty estimation.

public int DeepEnsembleSize { get; set; }

Property Value

int

Remarks

This value is only used when Method is DeepEnsemble.

DenormalizeUncertainty

Gets or sets whether the returned uncertainty should be denormalized to match the output scale.

public bool DenormalizeUncertainty { get; set; }

Property Value

bool

Remarks

When enabled, the library attempts to scale variances according to the output normalization transform. For non-linear normalization transforms, the uncertainty is returned in normalized space.

EnableIsotonicRegressionCalibration

Gets or sets whether to fit and apply isotonic regression calibration (binary calibration) when calibration labels are provided.

public bool EnableIsotonicRegressionCalibration { get; set; }

Property Value

bool

EnablePlattScaling

Gets or sets whether to fit and apply Platt scaling (binary calibration) when calibration labels are provided.

public bool EnablePlattScaling { get; set; }

Property Value

bool

EnableTemperatureScaling

Gets or sets whether to fit and apply temperature scaling for classification-like outputs when calibration labels are provided.

public bool EnableTemperatureScaling { get; set; }

Property Value

bool

Remarks

When enabled and calibration labels are provided via the builder, the system will calibrate predicted probabilities and return calibrated probabilities as the prediction output from uncertainty APIs.

Enabled

Gets or sets whether uncertainty quantification is enabled.

public bool Enabled { get; set; }

Property Value

bool

Remarks

When disabled, calls to uncertainty APIs fall back to deterministic behavior.

LaplacePriorPrecision

Gets or sets the prior precision (inverse variance) used by diagonal Laplace approximation.

public double LaplacePriorPrecision { get; set; }

Property Value

double

Method

Gets or sets the uncertainty quantification strategy to use.

public UncertaintyQuantificationMethod Method { get; set; }

Property Value

UncertaintyQuantificationMethod

MonteCarloDropoutRate

Gets or sets the dropout rate used when injecting Monte Carlo Dropout layers automatically.

public double MonteCarloDropoutRate { get; set; }

Property Value

double

Remarks

This value is only used when the model architecture does not already contain explicit MC dropout layers.

NumSamples

Gets or sets the number of stochastic samples to draw when using sampling-based methods.

public int NumSamples { get; set; }

Property Value

int

Remarks

Higher values generally improve estimate stability at the cost of increased inference latency.

PosteriorFitMaxSamples

Gets or sets the maximum number of samples used to fit Laplace/SWAG posteriors from calibration data.

public int PosteriorFitMaxSamples { get; set; }

Property Value

int

Remarks

This is a safety/performance bound to prevent extremely large calibration datasets from causing very slow builds.

RandomSeed

Gets or sets an optional random seed for reproducible Monte Carlo sampling.

public int? RandomSeed { get; set; }

Property Value

int?

SwagBurnInSteps

Gets or sets the number of initial SWAG steps to skip before collecting snapshots.

public int SwagBurnInSteps { get; set; }

Property Value

int

SwagLearningRate

Gets or sets the learning rate used for SWAG posterior fitting on calibration data.

public double SwagLearningRate { get; set; }

Property Value

double

SwagNumSnapshots

Gets or sets the number of SWAG snapshots to collect when fitting a SWAG posterior.

public int SwagNumSnapshots { get; set; }

Property Value

int

SwagNumSteps

Gets or sets the number of SWAG update steps used to collect snapshots.

public int SwagNumSteps { get; set; }

Property Value

int