Skip to contents

CATE score from a precomputed pseudo-outcome \(\rho\) (pure score; no weights)

Source: R/score-cate.R
score_cate_rho.Rd

Constructs a pure-score kernel for CATE experiments that uses only the precomputed pseudo-outcome \(\rho\). Per observation \(i\), the stored contribution is $$g_i \equiv \rho_i\,,$$ and for a subset \(S \subset \{1,\ldots,n\}\) of size \(s\) the kernel evaluates to the sum of contributions $$h(S) \;=\; \sum_{i \in S} g_i \;=\; \sum_{i \in S} \rho_i\,.$$

Usage

score_cate_rho(x0, x_cols, s, rho_col = NULL, rho = NULL)

Arguments

x0

Numeric length-\(p\) target point. Used only to check that length(x0) matches length(x_cols); it is not used in the kernel itself.

x_cols

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

s

Integer subset size (kernel order). Recorded in the score metadata; the pure kernel internally computes \(\sum_{i \in S} \rho_i\) for any subset \(S\) of size \(s\).

rho_col

Optional name of the column in data that contains the pseudo-outcome \(\rho\). If NULL, supply rho.

rho

Optional numeric vector (length \(n\)) of \(\rho\) values to use when rho_col is NULL.

Value

An ustat_score object with:

  • $name: a descriptive name,

  • $order: the kernel order \(s\),

  • $setup: stores g <- rho in the context,

  • $eval/$eval_batch: return \(\sum_{i \in S} g_i\) for each subset \(S\) (no weights applied).

Details

This score does not perform any nearest-neighbor ordering, kernel weighting, or L-stat aggregation. Any weighting/averaging of subset evaluations is handled outside the score layer (e.g., by the rank-/distance- weight modules). The arguments x0, x_cols, and s are accepted for API consistency and basic dimension checks; only \(\rho\) is used to construct the score.

Examples

demo <- cate_demo_data(n = 200, p = 2, design = "hetero", seed = 1)
sc <- score_cate_rho(
  x0 = rep(0, 2), x_cols = paste0("X", 1:2),
  s = 5, rho_col = "rho_oracle"
)
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) # three subset evaluations
#> [1] 1.9200414 0.8397658 4.4934999