Longitudinal Regression Trees and Forests

Builds regression trees and random forests for longitudinal or functional data using a spline projection method. Implements and extends the work of Yu and Lambert (1999) . This method allows trees and forests to be built while considering either level and shape or only shape of response trajectories.

splinetree

This package allows users to create, visualize, and evaluate regression trees and random forests for longitudinal or functional data through a spline projection method first suggested by Yu and Lambert (1999).

Installation

You can install splinetree from CRAN or from github with:

# install.packages("devtools")
devtools::install_github("anna-neufeld/splinetree")

Usage

Detailed information on using this package can be found in the package vignettes. The package vignettes can be accessed with:

browseVignettes(package='splinetree')

The vignettes are also available on the package website, https://anna-neufeld.github.io/splinetree/reference/index.html.

Example Tree

library(splinetree)
#> Loading required package: rpart
#> Loading required package: nlme
#> Loading required package: splines
tree1 <- splineTree(~HISP+WHITE+BLACK+HGC_MOTHER+HGC_FATHER+SEX+Num_sibs, 
        BMI ~ AGE, "ID", nlsySample, degree = 1, df=2, intercept = FALSE, cp = 0.005)
stPrint(tree1)
#> n= 1000,  
#> 
#> node), split, n , coefficients 
#>       * denotes terminal node
#> 
#> 1) root, 1000,  (4.961796, 8.091247) 
#>   2) WHITE< 0.5, 505,  (5.882807, 9.006190)*
#>   3) WHITE>=0.5, 495,  (4.022179, 7.157821) 
#>     6) HGC_FATHER< 8.5, 78,  (5.198284, 8.642817)*
#>     7) HGC_FATHER>=8.5, 417,  (3.802188, 6.880053)*

stPlot(tree)

Example Forest

set.seed(1234)
forest1 <- splineForest(~HISP+WHITE+BLACK+HGC_MOTHER+HGC_FATHER+SEX+Num_sibs, 
        BMI ~ AGE, "ID", nlsySample, degree = 1, df=2, intercept = FALSE, ntree=50, prob=1/2)
varImps <- varImpCoeff(forest1)