UncertaintyCalibrationData<TInput, TOutput>
Models & Types · AiDotNet.Models.Inputs
Provides optional calibration data for uncertainty quantification features.
For Beginners
Think of calibration data as a small “reality check” dataset. The model is trained on training data, then calibration data is used to tune uncertainty-related behavior without overfitting.
How It Works
Calibration data is used by certain uncertainty features that require a held-out dataset separate from training data. Examples include conformal prediction and probability calibration.
Properties
| Property | Summary |
|---|---|
TInput X { get; } |
Gets the calibration inputs. |
TOutput Y { get; } |
Gets the calibration targets (regression-style calibration). |
Vector<int> Labels { get; } |
Gets the calibration class labels (classification-style calibration). |
bool HasLabels { get; } |
Gets whether classification labels were provided. |
bool HasTargets { get; } |
Gets whether regression targets were provided. |
Methods
| Method | Summary |
|---|---|
UncertaintyCalibrationData<TInput, TOutput> ForClassification(TInput xCalibration, Vector<int> labels) |
Creates calibration data for classification-style calibration (e.g., temperature scaling and conformal prediction sets). |
UncertaintyCalibrationData<TInput, TOutput> ForRegression(TInput xCalibration, TOutput yCalibration) |
Creates calibration data for regression-style calibration (e.g., conformal prediction intervals). |