FunFor: Fit an FunFor model

In xiaotiand/FunFor: Random Forest for Functional Data (FunFor)

Description

This is the main function for functional forest.

Usage

FunFor(
  formula,
  data,
  mtry,
  ntree,
  importance = TRUE,
  npc = NULL,
  m_split = 10,
  smooth = TRUE,
  ...
)

Arguments

formula	Formula of model fitted
data	All the data
mtry	Variables tried at each split
ntree	Number of trees
importance	Whether to calculate PVIM
npc	A given number of PCs when smoothing
m_split	Optimal tree size
smooth	Whether to smooth curves

Details

The FunFor algorithm is able to predict curve responses for new observations and select important variables from a large set of scalar predictors. see ?optimal_size for how to determine the optimal size of a tree fit. See ?predict.mvRF for how to predict from new observations based on a fitted FunFor model.

Value

imp Importance measures for the candidate predictors

fits A list of the fitted trees. Will be used to make predictions.

Examples

library(MASS)
nbx = 100
nbobs = 100
T = seq(0, 1, len = 100)
m = 1
rho = 0.6
mu = matrix(1, nbx, 1)
ar1_cor = function(n, m,rho) {
 exponent = abs(matrix(1:n - 1, nrow = n, ncol = n, byrow = TRUE) - (1:n - 1))
 L = rho^exponent
 diag(L) = m
 L
}

p = nbx
x_sigma = ar1_cor(p, m, rho)
noise_sigma = ar1_cor(length(T), (5 * cos(T) + rnorm(length(T), 0, 1)) / 10, 0.01)
beta_1 = function(t) sin(20 * pi * T) / 3

X = mvrnorm(nbx, mu, x_sigma)
X = as.data.frame(X)
Y = data.frame(matrix(NA, nrow = nbobs, ncol = length(T)))
for(j in 1:nbobs) Y[j, ] = (X[j, 2] * X[j, 3]) * beta_1(T) + mvrnorm(1, rep(0, length(T)), noise_sigma)

formula = paste("data[, 1:length(T)]", "~", paste(names(X), collapse = "+"))
data = cbind(Y, X)
o_split = optimal_size(formula, data)
funfor_fit = FunFor(formula, data, mtry = 40, ntree = 10, npc = 3, m_split = o_split)

xiaotiand/FunFor documentation built on Dec. 23, 2021, 6:18 p.m.