Editare

loss functions: Classification and Regression Loss functions

  • Articol

The loss functions for classification and regression.

Usage

  expLoss(beta = 1, ...)

  hingeLoss(margin = 1, ...)

  logLoss(...)

  smoothHingeLoss(smoothingConst = 1, ...)

  poissonLoss(...)

  squaredLoss(...)

Arguments

beta

Specifies the numeric value of beta (dilation). The default value is 1.

margin

Specifies the numeric margin value. The default value is 1.

smoothingConst

Specifies the numeric value of the smoothing constant. The default value is 1.

...

hidden argument.

Details

A loss function measures the discrepancy between the prediction of a machine learning algorithm and the supervised output and represents the cost of being wrong.

The classification loss functions supported are:

logLoss

expLoss

hingeLoss

smoothHingeLoss

The regression loss functions supported are:

poissonLoss

squaredLoss.

Value

A character string defining the loss function.

Author(s)

Microsoft Corporation Microsoft Technical Support

See also

rxFastLinear, rxNeuralNet

Examples


 train <- function(lossFunction) {

     result <- rxFastLinear(isCase ~ age + parity + education + spontaneous + induced,
                   transforms = list(isCase = case == 1), lossFunction = lossFunction,
                   data = infert,
                   type = "binary")
     coef(result)[["age"]]
 }

 age <- list()
 age$LogLoss <- train(logLoss())
 age$LogLossHinge <- train(hingeLoss())
 age$LogLossSmoothHinge <- train(smoothHingeLoss())
 age