LearnerParam {ParamHelpers}

Create a description object for a parameter of a machine learning algorithm.

Description

This specializes Param() by adding a few more attributes, like a default value, whether it refers to a training or a predict function, etc. Note that you can set length to NA

The S3 class is a Param() which additionally stores these elements:

when character(1)

See argument of same name.

See the note in Param() about being able to pass expressions to certain arguments.

Usage

makeNumericLearnerParam(
  id,
  lower = -Inf,
  upper = Inf,
  allow.inf = FALSE,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeNumericVectorLearnerParam(
  id,
  len = as.integer(NA),
  lower = -Inf,
  upper = Inf,
  allow.inf = FALSE,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeIntegerLearnerParam(
  id,
  lower = -Inf,
  upper = Inf,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeIntegerVectorLearnerParam(
  id,
  len = as.integer(NA),
  lower = -Inf,
  upper = Inf,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeDiscreteLearnerParam(
  id,
  values,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeDiscreteVectorLearnerParam(
  id,
  len = as.integer(NA),
  values,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeLogicalLearnerParam(
  id,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeLogicalVectorLearnerParam(
  id,
  len = as.integer(NA),
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeUntypedLearnerParam(
  id,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

makeFunctionLearnerParam(
  id,
  default,
  when = "train",
  requires = NULL,
  tunable = TRUE,
  special.vals = list()
)

Arguments

id

(character(1))
Name of parameter.

lower

(numeric | expression)
Lower bounds. A singe value of length 1 is automatically replicated to len for vector parameters. If len = NA you can only pass length-1 scalars. Default is -Inf.

upper

(numeric | expression)
Upper bounds. A singe value of length 1 is automatically replicated to len for vector parameters. If len = NA you can only pass length-1 scalars. Default is Inf.

allow.inf

(logical(1))
Allow infinite values for numeric and numericvector params to be feasible settings. Default is FALSE.

default

(any concrete value | expression)
Default value used in learner. Note: When this is a discrete parameter make sure to use a VALUE here, not the NAME of the value. If this argument is missing, it means no default value is available.

when

(character(1))
Specifies when parameter is used in the learner: “train”, “predict” or “both”. Default is “train”.

requires

(NULL | call | expression)
States requirements on other parameters' values, so that setting this parameter only makes sense if its requirements are satisfied (dependent parameter). Can be an object created either with expression or quote, the former type is auto-converted into the later. Only really useful if the parameter is included in a (ParamSet()). Default is NULL which means no requirements.

tunable

(logical(1))
Is this parameter tunable? Defining a parameter to be not-tunable allows to mark arguments like, e.g., “verbose” or other purely technical stuff. Note that this flag is most likely not respected by optimizing procedures unless stated otherwise. Default is TRUE (except for untyped, function, character and characterVector) which means it is tunable.

special.vals

(list())
A list of special values the parameter can except which are outside of the defined range. Default is an empty list.

len

(integer(1))
Length of vector parameter. Can be set to NA to define a vector with unspecified length.

values

(vector | list | expression)
Possible discrete values. Instead of using a vector of atomic values, you are also allowed to pass a list of quite “complex” R objects, which are used as discrete choices. If you do the latter, the elements must be uniquely named, so that the names can be used as internal representations for the choice.

Value

LearnerParam().