predict.ITR: Treatment Prediction for rcDT and rcRF Models

In kdoub5ha/mvITR: Discovery of risk controlled individualized treatment rules (ITRs)

Description

Used to make treatment prediction for a rcDT and rcRF models. If the input is rcRF (forest), then the proportion of trees voting for treatment ('trt=1') is returned. If the input is rcDT (single tree), then the function returns the vote (0 / 1) for the model.

Usage

## S3 method for class 'ITR'
predict(fit, new.data, split.var, ctgs = NULL)

Arguments

fit	tree or forest object from 'grow.ITR' or 'Build.RF.ITR'.
new.data	data for which predictions are desired
split.var	splitting variables from the model fit
ctgs	columns of categorical variables.

Value

A summary list of the following elements:

SummaryTreat	proportion of trees voting for treatment (trt=1). If input is rcDT (single tree) then SummaryTreat is a single number. If input is rcRF (forest) then SummaryTreat is a vector equal to the length of the number of trees.
trt.pred	vector of treatment assignments 0, 1 based on the tree vote (single tree) or majority of tree votes (forest). This vector has length equal to the number of rows in 'new.data'.
n.trees	number of tree in 'fit'
tree.votes	matrix of votes for each tree for each subject in 'new.data'. Rows correspond to trees in 'fit' and columns correspond to subjects in 'new.dat'.
data	input data frame 'new.data'
NA.trees	number of trees returning no votes. In a forest, this is the number of null trees.

Examples

dat <- generateData()
# Generates rcDT using simualated data with splitting variables located in columns 1-10.
tree <- grow.ITR(data = dat, split.var = 1:10, 
                 risk.control = TRUE, risk.threshold = 2.75, 
                 lambda = 1)
# Predict treatment assignments for 1000 observations in `dat`
preds <- predict.ITR(fit = tree, new.data = dat, split.var = 1:10)

kdoub5ha/mvITR documentation built on April 7, 2020, 3:59 a.m.