MoTBFs (version 2.0)

motbf.fit: Learning hybrid BNs with MoTBFs

Description

Learn mixtures of truncated basis functions in a full hybrid network.

Usage

motbf.fit(
  graph,
  data,
  numIntervals = 4,
  POTENTIAL_TYPE = "MOP",
  maxParam = NULL,
  s = NULL,
  priorData = NULL,
  scale = TRUE
)
MoTBFs_Learning(
  graph,
  data,
  numIntervals = 4,
  POTENTIAL_TYPE = "MOP",
  maxParam = NULL,
  s = NULL,
  priorData = NULL,
  scale = TRUE
)

Value

A list of lists. Each list contains two elements

Child

A "charater" string which contains the name of the child variable.

functions

A list of three elements: the name of the parents; a "numeric" vector indicating the interval of the parent; and the fitted function in this interval.

Arguments

graph

A network of the class "bn" (from bnlearn package).

data

An object of class "data.frame"; it can contain continuous and discrete variables.

numIntervals

A positive integer indicating the maximum number of intervals for splitting the domain of the continuous parent variables. By default, it is set to 4.

POTENTIAL_TYPE

A "character" string, either MOP or MTE, corresponding to the type of basis function. By default, it is set to MOP.

maxParam

A positive integer which indicates the maximum number of coefficients in the function. If specified, the output is the function which gets the best BIC with, at most, this number of parameters. By default, it is set to NULL.

s

A "numeric" value which specifies the expert confidence in the prior knowledge. This argument takes values on the interval \([0, N]\), where \(N\) is the sample size, and is used to synchronize the support of the prior knowledge and the sample. By default, it is NULL, and must be modified only if prior information is to be incorporated to the fits.

priorData

An object of class "data.frame", corresponding to the prior information.

scale

A "logical" value indicating whether to standardize the numeric variables to have mean 0 and standard deviation 1.

Details

If the variable is discrete then it computes the probabilities and the size of each leaf. Children that have discrete parents have as many functions as configurations of the parents. Children that have continuous parents have as many functions as the number indicated in the argument "numIntervals" for each parent. Children that have mixed parents, combine both methods. The BIC criterion is used to decide the number of splitting points of the parent domains and to choose the number of basis functions used.

Examples

## Dataset Ecoli
require(MoTBFs)
data(ecoli)
data <- ecoli[,-c(1)] ## remove variable sequence

## Directed acyclic graph
dag <- LearningHC(data)

## Learning BN
intervals <- 3
potential <- "MOP"
bn1 <- motbf.fit(graph = dag, data = data, numIntervals = intervals, 
POTENTIAL_TYPE = potential, maxParam = 5)
bn1

## Learning BN
intervals <- 4
potential <- "MTE"
bn2 <- motbf.fit(graph = dag, data = data, numIntervals = intervals, 
POTENTIAL_TYPE = potential, maxParam = 15)
bn2