Package org.ddogleg.optimization.loss

Interface LossFunction

All Superinterfaces:
FunctionNtoS
All Known Implementing Classes:
LossCauchy.Function, LossHuber.Function, LossHuberSmooth.Function, LossIdentity.Function, LossIRLS, LossSquared.Function, LossTukey.Function, LossWeighted
public interface LossFunction extends FunctionNtoS

Residual loss function for regression. The standard squared loss can be sensitive to outliers. This function enables robust loss functions to be used instead which are less sensitive to. Most implementations will attempt to behave like squared error when an observation is not an outlier. Therefor, the following should be true f(0) = 0.

All implementations should return a scalar which is &ge 0. A value of zero would indicate no errors.

  • Method Summary

    Modifier and Type
    Method
    Description
    default boolean
    fixate(double[] residuals)
    Passes in the current residuals at the start of an iteration.
    int
    getNumberOfFunctions()
    Number of elements in the residual
    default int
    getNumOfInputsN()
    The number of inputs.
    void
    setNumberOfFunctions(int value)
     

    Methods inherited from interface org.ddogleg.optimization.functions.FunctionNtoS

    process

  • Method Details

    • fixate

      default boolean fixate(double[] residuals)
      Passes in the current residuals at the start of an iteration. If a loss function is dynamically computed and conditional on the residuals, here's where it should be done
      Returns:
      true if the loss function has changed and the cost needs to be recomputed.

    • getNumberOfFunctions

      int getNumberOfFunctions()
      Number of elements in the residual

    • setNumberOfFunctions

      void setNumberOfFunctions(int value)

    • getNumOfInputsN

      default int getNumOfInputsN()
      Description copied from interface: FunctionNtoS
      The number of inputs. Typically the parameters you are optimizing.
      Specified by:
      getNumOfInputsN in interface FunctionNtoS
      Returns:
      Number of inputs.