ucurve: U-curve algorith on the Partition Lattice Learning Space

In dmarcondes/partitionUcurve: U-curve Algorithm for Model Selection on the Partition Lattice Learning Space

Description

TBD

Usage

ucurve(
  xtrain,
  ytrain,
  xval,
  yval,
  optimal = T,
  exhaust = 1000,
  sampleNeigh = F,
  verbose = T,
  stop = 0,
  path = "~/GDrive/Doutorado/Códigos/Particoes/",
  Lh = NULL,
  cores = 4
)

Arguments

xtrain	Vector with the training sample of x values.
ytrain	Vector with the training sample of y values.
xval	Vector with the validation sample of x values.
yval	Vector with the validation sample of y values.
optimal	Logical indicating if the algorithm should return an optimal solution.
exhaust	Number of points to exhaust before stopping the algorithm.
sampleNeigh	Either false to consider all neighboors, or the maximum number of neighboors to sample at each exhaustion. If a number,then optimal should be false.
verbose	Logical to print a trace of the algorithm.
stop	Number of nodes yet to evaluate to trigger exhaustive search.
path	Path to preprocessed partition files.
Lh	A data frame with the partition lattice.
cores	Number of cores for parallel computing.

Details

TBD

Value

hypotheses	The estimated hypothesis of the global minimums with least VC dimension.
partitions	Partitions of the global minimums with least VC dimension.
error	Validation error of the global minimums.
exhausted	Number of nodes exhausted during algorithm.
remain	Number of nodes remaining after algorithm stopped.
finished	If the algorithm was finished or ended after not finding any Strong Local Minimum.
SLMvis	Number of nodes exhasuted until the last Strong Local Minimum was found.
remain_after_prune	Number of nodes remaining after finding each Strong Local Minimum.
exhausted_until_prune	Number of nodes exhausted until finding each Strong Local Minimum.
optimal	Wheter an optimal solution was returned.

Examples

set.seed(1)
x <- sample(x = c("01","02","03","04","05","06","07","08","09","10","11","12","13","14","15","16","17","18","19","20"),size = 500,replace = T)
y <- as.factor(ifelse(as.numeric(x)-10+rnorm(500,0,20/3) > 0,1,0))
x <- factor(x)
train <- sample(1:500,350,F)
xtrain <- x[train]
ytrain <- y[train]
xval <- x[!(c(1:500) %in% train)]
yval <- y[!(c(1:500) %in% train)]
ucurve(xtrain,ytrain,xval,yval,optimal = F,sampleNeigh = 5000)

dmarcondes/partitionUcurve documentation built on March 29, 2021, 10:23 p.m.