Counterfactual Explanation

Finds the minimal change in FPC scores needed to achieve a target prediction value. For logistic models, uses gradient-based optimization to find the class-flipping counterfactual.

Usage

fregre.counterfactual(
  model,
  data,
  observation,
  target.value = NULL,
  max.iter = 100,
  step.size = 0.1
)

Arguments

model

A fitted fregre.lm or fregre.logistic model.

data

An fdata object.

observation

Index of the observation (1-based).

target.value

Target prediction value (for lm models).

max.iter

Maximum iterations for logistic counterfactual (default 100).

step.size

Step size for logistic counterfactual (default 0.1).

Value

A list with observation, original_scores, counterfactual_scores, delta_scores, delta_function, distance, original_prediction, counterfactual_prediction, and found.

Examples

# \donttest{
fd <- fdata(matrix(rnorm(500), nrow = 50), argvals = seq(0, 1, length.out = 10))
y <- rnorm(50)
fit <- fregre.lm(fd, y, ncomp = 3)
result <- fregre.counterfactual(fit, fd, observation = 1, target.value = 0)
# }