rcDT.select: Optimal rcDT model selection

In kdoub5ha/rcITR: Risk Controlled ITR Discovery

Description

Performs k-fold cross validation for tuning of risk and tree size paramters to select the optimal rcDT model.

Usage

rcDT.select(data, split.var, N0 = 20, n0 = 5, efficacy = "y",
  risk = "r", col.trt = "trt", col.prtx = "prtx", lambda.seq = NA,
  lambda.length = 50, risk.control = TRUE, risk.threshold = NA,
  nfolds = 10, AIPWE = FALSE, sort = TRUE, ctg = NA,
  mtry = length(split.var), max.depth = 15, stabilize.type = c("linear",
  "rf"), stabilize = TRUE, use.other.nodes = TRUE, use.bootstrap = FALSE,
  extremeRandomized = FALSE, verbose = TRUE)

Arguments

data	data.frame. Data used to construct rcDT model. Must contain efficacy variable (y), risk variable (r), binary treatment indicator coded as 0 / 1 (trt), propensity score (prtx), candidate splitting covariates (split.var).
split.var	numeric vector. Columns of spliting variables.
N0	numeric specifying minimum number of observations required to call a node terminal. Defaults to 20.
n0	numeric specifying minimum number of treatment/control observations needed in a split to declare a node terminal. Defaults to 5.
efficacy	char. Efficacy outcome column. Defaults to 'y'.
risk	char. Risk outcome column. Defaults to 'r'.
col.trt	char. Treatment indicator column name. Should be of form 0/1 or -1/+1.
col.prtx	char. Propensity score column name.
lambda.seq	numeric vector. Identifies sequence of risk penalty parameters to be considered. Defaults to NA and will attempt to identify reasonable range.
lambda.length	numeric indicating number of risk penalty parameters to use in tuning. Larger values will cause model selection to be slower. Defaults to 50.
risk.control	logical. Should risk be controlled? Defaults to TRUE.
risk.threshold	numeric. Desired level of risk control.
AIPWE	logical. Should AIPWE (TRUE) or IPWE (FALSE) be used. Not available yet.
sort	internal use.
ctg	numeric vector corresponding to the categorical input columns. Defaults to NULL. Not available yet.
mtry	numeric specifying the number of randomly selected splitting variables to be included. Defaults to number of splitting variables.
max.depth	numeric specifying maximum depth of the tree. Defaults to 15 levels.
stabilize.type	character specifying method used for estimating residuals. Current options are 'linear' for linear model (default) and 'rf' for random forest.
stabilize	logical indicating if efficacy should be modeled using residuals. Defaults to TRUE.
use.other.nodes	logical. Should global estimator of objective function be used. Defaults to TRUE.
use.bootstrap	logical. Should a bootstrap resampling be done? Defaults to FALSE.
extremeRandomized	logical. Experimental for randomly selecting cutpoints in a random forest model. Defaults to FALSE and users should change this at their own peril.
verbose	logical. Should tuning progress bar be displayed. Defaults to TRUE.
n.folds	numeric. Number of folds to use in k-fold cross validation. Defaults to 10.
test	data.frame of testing observations. Should be formatted the same as 'data'.

Value

A summary of the cross validation including optimal penalty parameter and the optimal model.

best.tree	optimal rcDT model
alpha	tree size penalty
lambda	risk penalty
full.tree	unpruned tree
pruned.tree	output from pruning of 'full.tree'
subtrees	sequence of optimally pruned subtrees
best.tree.summaries	summary across trees
in.train	training samples from splits
in.test	testing samples from splits
elapsed.time	time elapsed during model tuning

Examples

# Grow large tree
set.seed(123)
dat <- generateData()
fit <- rcDT.select(data = dat, 
                   split.var = 1:10, 
                   nfolds = 5,
                   risk.threshold = 2.75)

kdoub5ha/rcITR documentation built on Aug. 5, 2020, 9:05 p.m.