LinearEvaluator<T>
Models & Types · AiDotNet.SelfSupervisedLearning.Evaluation
Linear evaluation protocol for assessing SSL representation quality.
For Beginners
Linear evaluation is the standard way to measure how good self-supervised representations are. We freeze the pretrained encoder and train only a simple linear classifier on top. Better representations = higher accuracy.
How It Works
Why linear evaluation?
Typical protocol:
Constructors
| Constructor | Summary |
|---|---|
LinearEvaluator(INeuralNetwork<T> encoder, int inputDim, int numClasses, double learningRate, int epochs) |
Initializes a new instance of the LinearEvaluator class. |
Properties
| Property | Summary |
|---|---|
int InputDimension { get; } |
Gets the input dimension (encoder output). |
int NumClasses { get; } |
Gets the number of classes for classification. |
Methods
| Method | Summary |
|---|---|
LinearEvalResult<T> Train(Tensor<T> trainData, int[] trainLabels, Tensor<T> validData, int[] validLabels) |
Trains the linear classifier on the given dataset. |
double Evaluate(Tensor<T> testData, int[] testLabels) |
Evaluates the linear classifier on a test dataset. |
double EvaluateTopK(Tensor<T> testData, int[] testLabels, int k) |
Computes top-k accuracy. |