package jsat.lossfunctions;
import jsat.classifiers.CategoricalResults;
import jsat.linear.Vec;
/**
* Specifies a loss function for multi-class problems. A multi-class loss
* function must support a raw vector of scores for each class, where positive
* values indicate preference for the class associated with the same index.
*
* Calling {@link #process(jsat.linear.Vec, jsat.linear.Vec) } on the raw
* scores is a mandatory first step, and will transform the raw scores into a
* usable form for the loss function.
*
* @author Edward Raff
*/
public interface LossMC extends LossC
{
/**
* Computes the scalar loss for on the given example
* @param processed the vector of raw predictions.
* @param y the true class label in [0, k-1] for k classes
* @return the loss in [0, Inf)
*/
public double getLoss(Vec processed, int y);
/**
* Given the vector of raw outputs for each class, transform it into a new
* vector.
*
* {@code processed} and {@code derivs} may be the same object, and will
* simply have all its values altered if so.
* @param pred the vector of raw predictions
* @param processed the location to store the processed predictions.
*/
public void process(Vec pred, Vec processed);
/**
* Computes the derivatives with respect to each output
*
* {@code processed} and {@code derivs} may be the same object, and will
* simply have all its values altered if so.
* @param processed the processed predictions
* @param derivs the vector to place the derivative of the loss to.
* @param y the true class label in [0, k-1] for k classes
*/
public void deriv(Vec processed, Vec derivs, int y);
/**
* Given the {@link #process(jsat.linear.Vec, jsat.linear.Vec) processed}
* predictions, returns the classification results for said predictions.
* @param processed the processed score/prediction vector
* @return the classification results
*/
public CategoricalResults getClassification(Vec processed);
}
Ads help maintain this website.