weighted.loss: Weighted Loss Function

In midr: Learning from Black-Box Models by Maximum Interpretation Decomposition

Weighted Loss Function

Description

weighted.loss() computes various loss metrics (e.g., RMSE, MAE) between two numeric vectors, or for the deviations from the weighted mean of a numeric vector.

Usage

weighted.loss(
  x,
  y = NULL,
  w = NULL,
  na.rm = FALSE,
  method = c("rmse", "mse", "mae", "medae", "r2")
)

Arguments

x	a numeric vector.
y	an optional numeric vector. If NULL, x is compared against its weighted mean.
w	a numeric vector of sample weights for each value in x.
na.rm	logical. If TRUE, any NA and NaNs are removed from all input vectors before the calculation.
method	the loss measure. One of "mse" (mean square error), "rmse" (root mean square error), mae" (mean absolute error), "medae" (median absolute error), or "r2" (R-squared).

Value

weighted.loss() returns a single numeric value.

Examples

# Calculate loss metrics between x and y with weights
weighted.loss(x = c(0, 10), y = c(0, 0), w = c(99, 1), method = "rmse")
weighted.loss(x = c(0, 10), y = c(0, 0), w = c(99, 1), method = "mae")
weighted.loss(x = c(0, 10), y = c(0, 0), w = c(99, 1), method = "medae")

# Verify uninterpreted variation ratio of a fitted MID model without weights
mid <- interpret(dist ~ speed, cars)
1 - weighted.loss(cars$dist, predict(mid, cars), method = "r2")
mid$ratio

# Verify uninterpreted variation ratio of a fitted MID model with weights
w <- 1:nrow(cars)
mid <- interpret(dist ~ speed, cars, weights = w)
1 - weighted.loss(cars$dist, predict(mid, cars), w = w, method = "r2")
mid$ratio

midr documentation built on Jan. 16, 2026, 5:06 p.m.