bartDiag function

bartDiag

Displays a selection of diagnostic plots for a BART model.

bartDiag(
  model,
  data,
  response,
  burnIn = 0,
  threshold = "Youden",
  pNorm = FALSE,
  showInterval = TRUE,
  combineFactors = FALSE
)

Arguments

• model: a model created from either the BART, modelarts, or bartMachine package.
• data: A dataframe used to build the model.
• response: The name of the response for the fit.
• burnIn: Trace plot will only show iterations above selected burn in value.
• threshold: A dashed line on some plots to indicate a chosen threshold value (classification only). by default the Youden index is shown.
• pNorm: apply pnorm to the y-hat data (classification only).
• showInterval: LOGICAL if TRUE then show 5% and 95% quantile intervals on ROC an PC curves (classification only).
• combineFactors: Whether or not to combine dummy variables (if present) in display.

Returns

A selection of diagnostic plots.

Examples

# For Regression
# Generate Friedman data
fData <- function(n = 200, sigma = 1.0, seed = 1701, nvar = 5) {
  set.seed(seed)
  x <- matrix(runif(n * nvar), n, nvar)
  colnames(x) <- paste0("x", 1:nvar)
  Ey <- 10 * sin(pi * x[, 1] * x[, 2]) + 20 * (x[, 3] - 0.5)^2 + 10 * x[, 4] + 5 * x[, 5]
  y <- rnorm(n, Ey, sigma)
  data <- as.data.frame(cbind(x, y))
  return(data)
}
f_data <- fData(nvar = 10)
x <- f_data[, 1:10]
y <- f_data$y

# Create dbarts model
library(dbarts)
set.seed(1701)
dbartModel <- bart(x, y, ntree = 5, keeptrees = TRUE, nskip = 10, ndpost = 10)

bartDiag(model = dbartModel, response = "y", burnIn = 100, data = f_data)

# For Classification
data(iris)
iris2 <- iris[51:150, ]
iris2$Species <- factor(iris2$Species)

# Create dbarts model
dbartModel <- bart(iris2[, 1:4], iris2[, 5], ntree = 5, keeptrees = TRUE, nskip = 10, ndpost = 10)

bartDiag(model = dbartModel, data = iris2, response = iris2$Species)