README

dartMachine

This is a fork of the bartMachine package with details listed below, designed for use in high-dimensional, sparse, settings. Of course, any bugs here are my own. For details, see Linero (2016). Bayesian Regression Trees for High Dimensional Prediction and Variable Selection. Journal of the American Statistical Association. To appear.

Instructions

For detailed instructions on installation, see the instructions on the bartMachine repository. This package is not available on CRAN, and must be installed by compiling from the source.

Illustration

The following illustration reproduces some components of Figure 4 from Linero (2016).

options(java.parameters = "-Xmx2g")
library(dartMachine)
    
Fried <- function(X) {
  10 * sin(pi * X[, 1] * X[,2]) + 20 * (X[,3] - 0.5)^2 + 10 * X[,4] + 5 * X[,5]
}
    
FriedSamp <- function(n, dim_x, sigma=1) {
  X <- matrix(runif(n * dim_x), n, dim_x)
  mu <- Fried(X)
  Y <- rnorm(n, mu, sigma)
    
  return(list(X=X, Y=Y))
}
    
RMSE <- function(x,y) sqrt(mean((x - y)^2))
    
set.seed(123)
train <- FriedSamp(250, 100, sqrt(10))
test  <- FriedSamp(1000, 100, sqrt(10))
    
## Fit bart
bart      <- bartMachine(X = as.data.frame(train$X), y = train$Y, seed = 1234, num_trees = 200)

## bartMachine initializing with 200 trees...
## Now building bartMachine for regression ...Covariate importance prior ON. 
## evaluating in sample data...done

bart_pred <- predict(bart, as.data.frame(test$X))
rm(bart); gc()

##           used (Mb) gc trigger  (Mb) max used  (Mb)
## Ncells 1493286 79.8    2637877 140.9  2637877 140.9
## Vcells 1660559 12.7   13674948 104.4 14020790 107.0

## Fit dart, with alpha = 1
dart   <- bartMachine(X = as.data.frame(train$X), y = train$Y,
                      do_ard = TRUE,
                      num_trees = 200,
                      num_burn_in = 5000, 
                          num_iterations_after_burn_in = 5000,
                      alpha_0 = 1, 
                            seed = 1234)

## bartMachine initializing with 200 trees...
## Now building bartMachine for regression ...Covariate importance prior ON. 
## evaluating in sample data...done

dart_pred    <- predict(dart, as.data.frame(test$X))
split_counts <- get_var_counts_over_chain(dart, type = "splits")
s_samples    <- get_cov_prior_select(dart)
rm(dart); gc()

##           used (Mb) gc trigger  (Mb) max used  (Mb)
## Ncells 1530046 81.8    2637877 140.9  2637877 140.9
## Vcells 2477540 19.0   13257030 101.2 17037812 130.0

## RMSE
RMSE(bart_pred, Fried(test$X))

## [1] 2.76089

RMSE(dart_pred, Fried(test$X))

## [1] 1.422885

## Reproduction of figure in paper
plot(colMeans(split_counts > 0), cex=.2)
abline(h = 0.5)

plot(colMeans(s_samples))