get_calibrated_prediction_regression_mondrian: Node-wise calibration of predictions of decision tree using...

In imbs-hl/timbR: Tree interpretation methods based on ranger

Node-wise calibration of predictions of decision tree using mondrian conformal prediction

Description

get_calibrated_prediction_regression_mondrian calibrates regression predictions using conformal prediction to a point prediction with prediction interval for each observation of the test data set.

Usage

get_calibrated_prediction_regression_mondrian(
  y_cal_pred,
  y_cal,
  y_test_pred,
  significance_level,
  tree,
  cal_data,
  test_data,
  dependent_varname,
  show_node_id = FALSE
)

Arguments

y_cal_pred	Predictions of decision tree for calibration set.
y_cal	True label of outcome variable for calibration set. Has to be in the same order as y_cal_pred.
y_test_pred	Predictions of test data set that should be calibrated.
significance_level	Level of uncertainty that should be reached by calibration, should be between 0 and 1.
tree	Tree whose predictions are to be calibrated, tree should be an object of class ranger.
cal_data	Data frame with calibration data (should contain same variables as train data).
test_data	Data frame with test data (should contain same variables as train data).
dependent_varname	Name of the dependent variable used to create the tree.
show_node_id	If true the ID of the terminal nodes of each prediction is returned as column in returned data frame.

Value

data frame with point prediction and lower and upper bound of prediction interval.

Author(s)

Lea Louisa Kronziel, M.Sc.

Examples

require(ranger)
require(timbR)
require(dplyr)

regr_data <- longley %>% data.frame()
# Train random forest with ranger
rf <- ranger(Employed ~ ., data = regr_data, num.trees = 10, importance = "permutation")

# Calculate pair-wise distances for all trees
rep_tree <- generate_tree(rf = rf, metric = "splitting variables", train_data = regr_data,
                                           dependent_varname = "Employed", importance.mode = TRUE, imp.num.var = 2)

# Get predictions
rep_tree_predictions <- predict(rep_tree, regr_data)$predictions

# Calibrated predictions
get_calibrated_prediction_regression_mondrian(y_cal_pred = rep_tree_predictions, y_cal = regr_data$Employed,
                                              y_test_pred = rep_tree_predictions, significance_level = 0.05,
                                              tree = rep_tree, cal_data = regr_data, test_data = regr_data,
                                              dependent_varname = "Employed")

imbs-hl/timbR documentation built on April 17, 2025, 2:08 p.m.