Extract, threshold, and intersect fixed-gate chain responses

Compose model-specific companion-importance extraction with activeByImportance() and an optional intersection over supplied preceding active sets. This convenience wrapper implements only the fixed threshold policy. chainPipeline() separately applies its fallback and common history policy around this result.

Usage

computeActiveByImportance(pathModel, suffix = ".o", thresh, stableK,
  history)

Arguments

pathModel: Directory passed to extractChainImportance().
suffix: Companion suffix passed to extraction and positional response name removal in activeByImportance().
thresh: Numeric inclusive mean normalized-importance threshold in percentage points. It has no default and is forwarded without validation.
stableK: Value compared with 1 and used as the number of current plus supplied preceding sets to intersect. It is not coerced or validated here.
history: List-like object whose final stableK - 1 elements are supplied to intersect() after the current raw set. The function does not establish whether these are raw or previously stabilized sets.

Value

A character vector of stripped response names, in current fixed-gate order, that survive the optional intersections. It can be empty.

Composition rule

Let $G_0$ be the character vector returned by activeByImportance(extractChainImportance(...), thresh = thresh). Extraction always uses its default useModel = TRUE. If stableK <= 1, or fewer than stableK - 1 history elements are available, the helper returns $G_0$ unchanged. Otherwise, for the retained history tail $G_1,\ldots,G_{K-1}$, it returns $$G_0\cap G_1\cap\cdots\cap G_{K-1}.$$ Base intersect() preserves the current vector's order while removing duplicate names. No fallback replaces an empty extraction or intersection.

In chainPipeline(), this wrapper is called with stableK = 1 and an empty history. The pipeline first replaces an empty raw result with all declared responses, then applies its own window by intersecting the current set with preceding returned active sets. Consequently, standalone wrapper output and pipeline fallback/stability must not be described as one rule.

Conditions and side effects

Validation and failures are inherited from extraction, the fixed table gate, tail(), Reduce(), and intersect(). In particular, a missing stableK comparison can error, and unusual history element types retain base R behavior. The helper reads models through the extractor but writes no file, sets no seed, draws no random value directly, and emits no deliberate condition.

Examples

root <- tempfile("ssel-chain-stability-")
dir.create(root)
train <- data.frame(
  alpha.o = 1:12,
  beta.o = rep(c(0, 1), 6),
  x = rep(1:4, 3),
  y = c(2.1, 1.7, 3.5, 2.8, 5.2, 4.1,
        6.4, 5.3, 7.1, 6.8, 8.2, 7.5)
)
fit <- caret::train(
  y ~ ., data = train, method = "lm",
  trControl = caret::trainControl(method = "none")
)
saveRDS(fit, file.path(root, "lm_target_demo.Rds"))

computeActiveByImportance(
  pathModel = root,
  suffix = ".o",
  thresh = 0,
  stableK = 2,
  history = list("alpha")
)
#> [1] "alpha"
# Only a currently active name also present in the supplied set survives.

unlink(root, recursive = TRUE)