Skip to contents

Loss function

LossFunction.Rd

Represents a loss function for regression or classification tasks.

Details

Loss functions are expected to be vectorized and to accept a single numeric argument representing the input on the chosen scale:

  • "score": residuals y_true - y_pred for regression, or signed margins y_true * f(x) for binary classification.

  • "probability": predicted probabilities for the positive class. Visualizers obtain the loss for y = 0 by evaluating the function at 1 - p.

The returned value must be a numeric vector of losses with the same length as the input. Additional optional parameters can be encoded via closures and are configurable through lss(), which wraps the stored functions with the supplied arguments.

Public fields

id

(character(1))
Identifier for the loss function.

label

(character(1))
Label for the loss function.

task_type

(character(1))
Task type: "regr" or "classif".

input_default

(character(1))
The natural input scale for the loss: "score" or "probability".

input_supported

(character())
All input scales this loss can be expressed on (subset of c("score", "probability")).

fun

(function or list)
The loss function itself. Provide either a vectorized closure matching the active input scale or a named list of such closures keyed by "score" and / or "probability".

Methods

Public methods

Method new()

Creates a new instance of this R6 class.

Usage

LossFunction$new(
  id,
  label,
  task_type,
  fun,
  input_default = "score",
  input_supported = c("score")
)

Arguments

id

(character(1))
Identifier for the loss function.

label

(character(1))
Label for the loss function.

task_type

(character(1))
Task type: "regr" or "classif".

fun

(function or list)
The loss function. Supply a vectorized closure matching the active input scale or a named list of closures keyed by "score" / "probability".

input_default

(character(1))
Default input scale ("score" or "probability").

input_supported

(character())
Character vector of supported input scales. Must contain input_default.

Method get_fun()

Get the loss function for a specific input type.

Usage

LossFunction$get_fun(input_type = self$input_default)

Arguments

input_type

(character(1))
The input type: "score" or "probability".

Returns

A function for the specified input type.

Method clone()

The objects of this class are cloneable with this method.

Usage

LossFunction$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.