Score specimens with a frozen trimmed-rCLR signed panel — score_ws_rclr

Applies ws_rclr_trimmed to each row of the entire supplied matrix before selecting the frozen panel and summing its signed values. Background columns can change the centering set and the default specimen-specific pseudocount, so adding or removing them can change scores even when every panel feature is present. Use a consistent feature universe for comparable scores. Missing panel features always fail closed.

Usage

score_ws_rclr_panel(model, X, meta = NULL, strict = FALSE)

Arguments

model: A ws_rclr_panel_model from fit_ws_rclr_panel.
X: Numeric specimens-by-features matrix.
meta: Unused; accepted for the canonical roster score contract.
strict: Logical scalar. The default FALSE preserves the supplied- matrix transform and permits missing non-panel training columns for backward compatibility. Set TRUE to reject any missing column from the frozen training feature universe, including background columns. Extra columns are still transformed in both modes; strict mode checks absence only.

Value

One finite numeric score per specimen, with a support data-frame attribute containing one row per score. n_available counts training- universe columns present, n_fitted counts all training-universe columns, and full_support reports whether these counts agree. These are feature counts, not pair counts or the number of features retained for centering. They describe column availability, not detection or assay quality. A TRUE flag does not rule out effects of extra background columns. Save attr(score, "support") before subsetting or numeric coercion, which can discard attributes.

Examples

set.seed(7)
X <- matrix(rexp(20 * 8), 20, dimnames = list(NULL, paste0("f", 1:8)))
model <- fit_ws_rclr_panel(X, rep(0:1, each = 10),
                          hp = list(panel_size = 1, min_centering_size = 1,
                                    trim_upper = 0, trim_lower = 0))
full <- score_ws_rclr_panel(model, X, strict = TRUE)
panel_only <- X[, model$panel, drop = FALSE]
partial <- score_ws_rclr_panel(model, panel_only)
attr(partial, "support")  # 1 available / 8 fitted features; full_support = FALSE
#>    n_available n_fitted full_support
#> 1            1        8        FALSE
#> 2            1        8        FALSE
#> 3            1        8        FALSE
#> 4            1        8        FALSE
#> 5            1        8        FALSE
#> 6            1        8        FALSE
#> 7            1        8        FALSE
#> 8            1        8        FALSE
#> 9            1        8        FALSE
#> 10           1        8        FALSE
#> 11           1        8        FALSE
#> 12           1        8        FALSE
#> 13           1        8        FALSE
#> 14           1        8        FALSE
#> 15           1        8        FALSE
#> 16           1        8        FALSE
#> 17           1        8        FALSE
#> 18           1        8        FALSE
#> 19           1        8        FALSE
#> 20           1        8        FALSE
stopifnot(any(abs(as.numeric(full) - as.numeric(partial)) > 1e-8))
tryCatch(score_ws_rclr_panel(model, panel_only, strict = TRUE),
         singlesample_input_error = function(e) e$code)
#> [1] "MISSING_REQUIRED_FEATURES"