Class AlignmentFeedbackData<T>
Contains human feedback data for AI alignment.
public class AlignmentFeedbackData<T>
Type Parameters
TThe numeric data type used for calculations.
- Inheritance
-
AlignmentFeedbackData<T>
- Inherited Members
Properties
Inputs
Gets or sets the input prompts or examples.
public Matrix<T> Inputs { get; set; }
Property Value
- Matrix<T>
Outputs
Gets or sets the model outputs for each input.
public Matrix<T> Outputs { get; set; }
Property Value
- Matrix<T>
Preferences
Gets or sets human preference comparisons.
public (int preferred, int notPreferred)[] Preferences { get; set; }
Property Value
- (int preferred, int notPreferred)[]
Remarks
Each element is a pair of indices into Outputs, with the first being preferred.
Ratings
Gets or sets numerical ratings for each output (optional).
public double[] Ratings { get; set; }
Property Value
- double[]
Rewards
Gets or sets reward labels for reinforcement learning.
public double[] Rewards { get; set; }
Property Value
- double[]
TextualFeedback
Gets or sets textual feedback for outputs (optional).
public string[] TextualFeedback { get; set; }
Property Value
- string[]
Methods
EnsurePreferencesValid()
Validates preference indices and throws if invalid.
public void EnsurePreferencesValid()
Exceptions
- InvalidOperationException
Thrown when preference indices are out of bounds.
ValidatePreferences()
Validates that preference indices are within valid bounds.
public bool ValidatePreferences()
Returns
- bool
True if all preference indices are valid, false otherwise.
Remarks
For Beginners: This method checks that all the preference pairs reference valid output indices. Since preferences are pairs of indices pointing to specific outputs, they must all be within the range of available outputs.