Extract a tidy weight-profile frame from a fitted RdistNN object

Returns one row per (target point, rank) with the DNN weight, cumulative weight share, the originating observation index, and — when distances were stored — the distance to the target. Ranks are assigned by ascending stored distance when dx is available and by descending weight otherwise; the two orderings coincide for monotone single-scale DNN weights, while two-scale (TDNN) fits should request return = c("W", "dx") for distance-based ranks. For exact DNN fits the finite-population support edge is rank n - s + k, not rank s; s is the subsample/localization order.

Usage

plot_data_weights(object)

Arguments

object

A fitted result object holding the weight matrix W (stored by default via return = c("W")).

Value

A data.frame with columns target_id (integer), rank (integer, 1..n), obs_id (integer row index into the training data), weight, cum_share (cumulative weight share in rank order), and distance (NA when distances were not stored).

Examples

set.seed(1)
n <- 60
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, s = 10L, se = FALSE, return = c("W", "dx")
)
head(plot_data_weights(fit))
#>   target_id rank obs_id     weight cum_share  distance
#> 1         1    1      2 0.16666667 0.1666667 0.1877888
#> 2         1    2     60 0.14124294 0.3079096 0.2229032
#> 3         1    3     42 0.11932593 0.4272355 0.2568383
#> 4         1    4     16 0.10048499 0.5277205 0.2948897
#> 5         1    5     26 0.08433562 0.6120561 0.3376483
#> 6         1    6     49 0.07053525 0.6825914 0.4002750