mc.wbart.gse: Global SE variable selection for BART with parallel computation

Description

Here we implement the global SE method for variable selection in nonparametric survival analysis with BART. Unfortunately, the method is very computationally intensive so we present some trade-offs below.

Usage

mc.wbart.gse( x.train, y.train,
              P=50L, R=5L, ntree=20L, numcut=100L, C=1, alpha=0.05,
              k=2.0, power=2.0, base=0.95,
              ndpost=2000L, nskip=100L,
              printevery=100L, keepevery=1L, keeptrainfits=FALSE,
              seed=99L, mc.cores=2L, nice=19L 
              )

Arguments

x.train

Explanatory variables for training (in sample) data. Must be a matrix with (as usual) rows corresponding to observations and columns to variables. surv.bart will generate draws of \(f(t, x)\) for each \(x\) which is a row of x.train.

y.train

The continuous outcome.

The number of permutations: typically 50 or 100.

The number of replicates: typically 5 or 10.

ntree

The number of trees. In variable selection, the number of trees is smaller than what might be used for the best fit.

numcut

The number of possible values of c (see usequants). If a single number if given, this is used for all variables. Otherwise a vector with length equal to ncol(x.train) is required, where the \(i^{th}\) element gives the number of c used for the \(i^{th}\) variable in x.train. If usequants is false, numcut equally spaced cutoffs are used covering the range of values in the corresponding column of x.train. If usequants is true, then min(numcut, the number of unique values in the corresponding columns of x.train - 1) c values are used.

The starting value for the multiple of SE. You should not need to change this except in rare circumstances.

alpha

The global SE method relies on simultaneous 1-alpha coverage across the permutations for all predictor variables.

k is the number of prior standard deviations \(f(t, x)\) is away from +/-3. The bigger k is, the more conservative the fitting will be.

power

Power parameter for tree prior.

base

Base parameter for tree prior.

ndpost

The number of posterior draws after burn in. In the global SE method, generally, the method is repeated several times to establish the variable count probabilities. However, we take the alternative approach of simply running the MCMC chain longer which should result in the same stabilization of the estimates. Therefore, the number of posterior draws in variable selection should be set to a larger value than would be typically anticipated for fitting.

nskip

Number of MCMC iterations to be treated as burn in.

printevery

As the MCMC runs, a message is printed every printevery draws.

keepevery

Every keepevery draw is kept.

keeptrainfits

If TRUE the draws of \(f(t, x)\) for \(x\) = rows of x.train are generated.

seed

seed required for reproducible MCMC.

mc.cores

Number of cores to employ in parallel.

nice

Set the job priority. The default priority is 19: priorities go from 0 (highest) to 19 (lowest).

Value

mc.wbart.gse returns a list.

References

Bleich, J., Kapelner, A., George, E.I., and Jensen, S.T. (2014). Variable selection for BART: an application to gene regulation. The Annals of Applied Statistics, 8:1750-81.

Examples

Run this code

# NOT RUN {
library(ElemStatLearn)

data(phoneme)

x.train <- matrix(NA, nrow=4509, ncol=257)
    
dimnames(x.train)[[2]] <- c(paste0('x.', 1:256), 'speaker')
    
x.train[ , 257] <- as.numeric(phoneme$speaker)

for(j in 1:256) x.train[ , j] <- as.numeric(phoneme[ , paste0('x.', j)])

gse <- mc.wbart.gse(x.train, as.numeric(phoneme$g), mc.cores=5, seed=99)

## important variables
dimnames(x.train)[[2]][gse$which]
    
# }