Quantile Regression Learner with SE Prediction

Wraps a quantile regression learner and converts quantile predictions to SE. Assumes approximate normality to map inter-quantile range to standard error.

Creates a new instance of this R6 class.

Arguments

learner

(mlr3::LearnerRegr)
Base quantile learner. Must support predict_type = "quantiles".

Details

This learner:

1. Trains a base learner that supports quantile predictions

2. predicts lower and upper quantiles

3. SE prediction is the inter-quantile range multiplied by a given factor

Fields

wrapped

(LearnerRegr)\cr Read-only access to the wrapped base learner.

param_set

(paradox::ParamSet)\cr The combined parameter set.

Parameters

• quantile_response :: numeric(1)
  Quantile response to use for the prediction. Initialized to 0.5 (median).

• quantile_lower :: numeric(1)
  Lower quantile for SE estimation. Initialized to 0.1 (10th percentile).

• quantile_upper :: numeric(1)
  Upper quantile for SE estimation. Initialized to 0.9 (90th percentile).

• se_factor :: numeric(1)
  Factor to multiply the inter-quantile range to get the SE. Initialized to 0.5.

The initial setup forwards the wrapped learner's response prediction and calculates the SE as half the range from predicted 0.1 to 0.9 quantiles.

Fields

• $wrapped :: mlr3::LearnerRegr\cr Read-only access to the wrapped base learner.

Examples

if (FALSE) { # \dontrun{
# Requires a quantile regression learner (e.g., from mlr3extralearners)
# base_learner <- lrn("regr.ranger")  # Assuming it supports quantiles
# learner <- lrn("regr.quantile_se", learner = base_learner)
# learner$param_set$set_values(quantile_lower = 0.1, quantile_upper = 0.9)

# Train and predict
# task <- tsk("mtcars")
# learner$train(task)
# pred <- learner$predict(task)
# pred$se  # Standard errors derived from quantiles
} # }