Skip to contents

PLQR diagnostic score (component \(j\)); pure score (no weights, no fitting) (experimental)

Source: R/score-plqr.R
score_plqr_component.Rd

Constructs a pure-score kernel for PLQR diagnostics that stores the per-observation score contributions $$ \psi_{i} \;=\; \Bigl(\tau - \mathbf{1}\{\,Y_{i} \le D_{i}^{\top}\theta + g_{0}(U_{i})\,\}\Bigr)\, \bigl(D_{i,j} - m_{D,ij}(U_{i})\bigr)\,, $$ and returns subset sums \(\sum_{i \in S} \psi_i\) in eval/eval_batch. This score performs no nearest-neighbor ordering or L-stat weighting; any aggregation/weights are handled outside the score layer.

Usage

score_plqr_component(
  x0,
  x_cols,
  D_cols,
  g_col,
  mD_cols,
  y_col,
  theta,
  j = 1L,
  tau = 0.5,
  s
)

Arguments

x0

Numeric length-\(p\) target point. Used only for basic dimension checks against x_cols; not used in the kernel.

x_cols

Integer or character indices of the \(X\) columns in data; used for dimension checks with x0.

D_cols

Integer or character indices of the \(D\) columns in data (length \(p_d\)).

g_col

Character; name of the column in data holding the shape-nuisance values \(g_0(U)\).

mD_cols

Integer or character indices of the \(m_{D}(U)\) columns in data; must have the same length and order as D_cols. The \(j\)-th column represents \(m_{D,ij}(U)\).

y_col

Name or index of the outcome column in data.

theta

Numeric vector of length length(D_cols) giving the fixed coefficient vector \(\theta\).

j

Integer component index in \(1,\ldots,\mathrm{length}(D\_cols)\).

tau

Quantile level in \((0,1)\).

s

Integer kernel order (subset size). Recorded in the score metadata; not used internally by this pure score.

Value

An ustat_score object whose $setup stores the vector score of contributions in the context, and whose $eval/$eval_batch return subset sums of score (no weights applied).

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

demo <- plqr_demo_data(n = 200, p = 3, p_d = 2, tau = 0.5, seed = 1)
sc <- score_plqr_component(
  x0      = rep(0, 3),
  x_cols  = paste0("X", 1:3),
  D_cols  = paste0("D", 1:2),
  g_col   = "g_oracle",
  mD_cols = paste0("mD", 1:2),
  y_col   = "Y",
  theta   = demo$truth$theta_true,
  j       = 1L,
  tau     = 0.5,
  s       = 5
)
ctx <- score_setup(sc, demo$data)
subs <- cate_demo_subsets(n = nrow(demo$data), K = 3, s = 5, seed = 1)
score_eval_batch(sc, subs, demo$data, ctx)
#> [1]  0.6484292 -0.1609615 -0.7863173