Package smile.base.cart

Interface Loss

public interface Loss
Regression loss function.

  • Nested Class Summary

    Nested Classes
    Modifier and Type
    Interface
    Description
    static enum 
    Loss.Type
    The type of loss.

  • Method Summary

    Modifier and Type
    Method
    Description
    static Loss
    huber(double p)
    Huber loss function for M-regression, which attempts resistance to long-tailed error distributions and outliers while maintaining high efficiency for normally distributed errors.
    double
    intercept(double[] y)
    Returns the intercept of model.
    static Loss
    lad()
    Least absolute deviation regression loss.
    static Loss
    logistic(int[] labels)
    Logistic regression loss for binary classification.
    static Loss
    logistic(int c, int k, int[] labels, double[][] p)
    Logistic regression loss for multi-class classification.
    static Loss
    ls()
    Least squares regression loss.
    static Loss
    ls(double[] y)
    Least squares regression loss.
    double
    output(int[] nodeSamples, int[] sampleCount)
    Calculate the node output.
    static Loss
    quantile(double p)
    Quantile regression loss.
    double[]
    residual()
    Returns the residual vector.
    double[]
    response()
    Returns the response variable for next iteration.
    static Loss
    valueOf(String s)
    Parses the loss.

  • Method Details

    • output

      double output(int[] nodeSamples, int[] sampleCount)
      Calculate the node output.
      Parameters:
      nodeSamples - the index of node samples to their original locations in training dataset.
      sampleCount - samples[i] is the number of sampling of dataset[i]. 0 means that the datum is not included and values of greater than 1 are possible because of sampling with replacement.
      Returns:
      the node output

    • intercept

      double intercept(double[] y)
      Returns the intercept of model.
      Parameters:
      y - the response variable.
      Returns:
      the intercept of model.

    • response

      double[] response()
      Returns the response variable for next iteration.
      Returns:
      the response variable for next iteration.

    • residual

      double[] residual()
      Returns the residual vector.
      Returns:
      the residual vector.

    • ls

      static Loss ls()
      Least squares regression loss. Least-squares is highly efficient for normally distributed errors but is prone to long tails and outliers.
      Returns:
      the least square regression loss.

    • ls

      static Loss ls(double[] y)
      Least squares regression loss. Least-squares is highly efficient for normally distributed errors but is prone to long tails and outliers.
      Parameters:
      y - the response variable.
      Returns:
      the least square regression loss.

    • quantile

      static Loss quantile(double p)
      Quantile regression loss. The gradient tree boosting based on this loss function is highly robust. The trees use only order information on the input variables and the pseudo-response has only two values {-1, +1}. The line searches (terminal node values) use only specified quantile ratio.
      Parameters:
      p - the percentile.
      Returns:
      the quantile regression loss.

    • lad

      static Loss lad()
      Least absolute deviation regression loss. The gradient tree boosting based on this loss function is highly robust. The trees use only order information on the input variables and the pseudo-response has only two values {-1, +1}. The line searches (terminal node values) use only medians. This is a special case of quantile regression of q = 0.5.
      Returns:
      the least absolute deviation regression loss.

    • huber

      static Loss huber(double p)
      Huber loss function for M-regression, which attempts resistance to long-tailed error distributions and outliers while maintaining high efficiency for normally distributed errors.
      Parameters:
      p - of residuals
      Returns:
      the Huber loss.

    • logistic

      static Loss logistic(int[] labels)
      Logistic regression loss for binary classification.
      Parameters:
      labels - the class label encoder.
      Returns:
      the logistic regression loss for binary classification.

    • logistic

      static Loss logistic(int c, int k, int[] labels, double[][] p)
      Logistic regression loss for multi-class classification.
      Parameters:
      c - the class id that this loss function fits on.
      k - the number of classes.
      labels - the class label encoder.
      p - the posteriori probabilities.
      Returns:
      the logistic regression loss for multi-class classification.

    • valueOf

      static Loss valueOf(String s)
      Parses the loss.
      Parameters:
      s - the string specification of loss.
      Returns:
      the loss function.