Cross-validation for Gradient Boosted Trees

Cross-validation for Gradient Boosted Trees

cv.gbt(
  object,
  K = 5,
  repeats = 1,
  params = list(),
  nrounds = 500,
  early_stopping_rounds = 10,
  nthread = 12,
  train = NULL,
  type = "classification",
  trace = TRUE,
  seed = 1234,
  maximize = NULL,
  fun,
  ...
)

Arguments

object	Object of type "gbt" or "ranger"
K	Number of cross validation passes to use (aka nfold)
repeats	Repeated cross validation
params	List of parameters (see XGBoost documentation)
nrounds	Number of trees to create
early_stopping_rounds	Early stopping rule
nthread	Number of parallel threads to use. Defaults to 12 if available
train	An optional xgb.DMatrix object containing the original training data. Not needed when using Radiant's gbt function
type	Model type ("classification" or "regression")
trace	Print progress
seed	Random seed to use as the starting point
maximize	When a custom function is used, xgb.cv requires the user indicate if the function output should be maximized (TRUE) or minimized (FALSE)
fun	Function to use for model evaluation (i.e., auc for classification and RMSE for regression)
...	Additional arguments to be passed to 'fun'

Value

A data.frame sorted by the mean of the performance metric

Details

See https://radiant-rstats.github.io/docs/model/gbt.html for an example in Radiant

See also

gbt to generate an initial model that can be passed to cv.gbt

Rsq to calculate an R-squared measure for a regression

RMSE to calculate the Root Mean Squared Error for a regression

MAE to calculate the Mean Absolute Error for a regression

auc to calculate the area under the ROC curve for classification

profit to calculate profits for classification at a cost/margin threshold

Examples

if (FALSE) {
result <- gbt(dvd, "buy", c("coupon", "purch", "last"))
cv.gbt(result, params = list(max_depth = 1:6))
cv.gbt(result, params = list(max_depth = 1:6), fun = "logloss")
cv.gbt(
  result,
  params = list(learning_rate = seq(0.1, 1.0, 0.1)),
  maximize = TRUE, fun = profit, cost = 1, margin = 5
)
result <- gbt(diamonds, "price", c("carat", "color", "clarity"), type = "regression")
cv.gbt(result, params = list(max_depth = 1:2, min_child_weight = 1:2))
cv.gbt(result, params = list(learning_rate = seq(0.1, 0.5, 0.1)), fun = Rsq, maximize = TRUE)
cv.gbt(result, params = list(learning_rate = seq(0.1, 0.5, 0.1)), fun = MAE, maximize = FALSE)
rig_wrap <- function(preds, dtrain) {
  labels <- xgboost::getinfo(dtrain, "label")
  value <- rig(preds, labels, lev = 1)
  list(metric = "rig", value = value)
}
result <- gbt(titanic, "survived", c("pclass", "sex"), eval_metric = rig_wrap, maximize = TRUE)
cv.gbt(result, params = list(learning_rate = seq(0.1, 0.5, 0.1)))
}