Plot estimate sensitivity to the kernel order

Evaluates or reuses the candidate kernel orders of a tuning grid (default: the paper rule's lambda grid via tuning_dnn_paper(), always including the fitted s) and plots the estimates with confidence bounds against s, marking the fitted kernel order with a filled point and a dotted guide. For grid fits, the stored tuning_grid payload is consumed directly. For scalar fits, the internal grid engine is used on the stored model state; public estimators are not refit.

Usage

plot_tuning(
  object,
  tuning = tuning_dnn_paper(),
  targets = NULL,
  level = 0.95,
  se = TRUE,
  candidates = "all",
  ...
)

# Default S3 method
plot_tuning(
  object,
  tuning = tuning_dnn_paper(),
  targets = NULL,
  level = 0.95,
  se = TRUE,
  candidates = "all",
  ...
)

# S3 method for class 'np_result'
plot_tuning(
  object,
  tuning = tuning_dnn_paper(),
  targets = NULL,
  level = 0.95,
  se = TRUE,
  candidates = "all",
  ...
)

# S3 method for class 'cate_result'
plot_tuning(
  object,
  tuning = tuning_dnn_paper(),
  targets = NULL,
  level = 0.95,
  se = TRUE,
  candidates = "all",
  ...
)

# S3 method for class 'plqr_result'
plot_tuning(
  object,
  tuning = tuning_dnn_paper(),
  targets = NULL,
  level = 0.95,
  se = TRUE,
  candidates = "all",
  ...
)

Arguments

object: A fitted np_result, cate_result, or plqr_result.
tuning: A tuning specification for the candidate grid (default tuning_dnn_paper()).
targets: Optional integer vector of target ids (defaults to at most three: first, middle, last).
level: Confidence level for the bounds (default 0.95).
se: Logical; compute jackknife bounds for each candidate when object is a scalar fit (default TRUE). Stored grid fits reuse the standard errors already present in object$tuning_grid.
candidates: Candidate rows to display. The default "all" shows the tuning sweep; a character vector of candidate ids or "default" can be used to restrict the display.
...: Ignored; reserved for future options.

Value

A ggplot object.

Details

Scalar grid evaluation uses the training data stored on the fit; fits from package versions predating that storage must be re-estimated first.

Experimental methods

Methods for plqr_result objects are experimental, as is all PLQR functionality in this release; see plqr_dnn().

Examples

# \donttest{
if (requireNamespace("ggplot2", quietly = TRUE)) {
  set.seed(1)
  n <- 100
  x_mat <- matrix(rnorm(n * 2), n, 2)
  y_vec <- x_mat[, 1] + rnorm(n, sd = 0.3)
  fit <- np_dnn(X = x_mat, x0 = c(0, 0), Y = y_vec, k = 1L)
  plot_tuning(fit)
}

# }