Plot local design conditioning for PLQR fits (experimental)

Two facets across targets: the smallest eigenvalue of the weighted residualized-design Gram matrix (linear scale, guide at zero) and its condition number (log10 scale, dashed guides at cond_limits, labelled caution and ill-conditioned). This is the software face of the local design-conditioning diagnostics: a well-conditioned local Gram is what makes the moment evaluation (and later inversion) trustworthy.

Usage

plot_conditioning(object, cond_limits = c(1000, 1e+06), ...)

Arguments

object

A plqr_result from plqr_dnn().

cond_limits

Numeric length-2 vector (default c(1e3, 1e6)); dashed guides drawn at these condition numbers.

...

Ignored; reserved for future options.

Value

A ggplot object.

Lifecycle

Experimental. PLQR functionality is outside the CATE and regression workflow focus of the initial public release. Its API and inference interfaces may change in later releases.

Examples

if (requireNamespace("ggplot2", quietly = TRUE)) {
  demo <- plqr_demo_data(n = 90, p = 2, p_d = 2, tau = 0.5, seed = 1)
  df <- demo$data
  fit <- plqr_dnn(
    X = as.matrix(df[paste0("X", 1:2)]),
    D = as.matrix(df[paste0("D", 1:2)]),
    g = as.numeric(df$g_oracle),
    mD = as.matrix(df[paste0("mD", 1:2)]),
    Y = as.numeric(df$Y),
    theta_eval = demo$truth$theta_true, tau = 0.5,
    X0 = rbind(c(0, 0), c(0.5, 0.5)), k = 1L, s = 8L, se = FALSE
  )
  plot_conditioning(fit)
}