Classify a single column under deployment rules (sample + gate + score)

The deployment entry point for one column. Draws a single row sample, runs the block-gated feature extractor and the value/name detector over it, and returns a prediction plus the feature row. This is the only place that should call fx_extract_features() in production, because it is the only place that guarantees the extractor never runs un-sampled.

Usage

fx_classify_column(
  x,
  name = NULL,
  n_rows = 500L,
  blocks = NULL,
  features_used = NULL,
  manifest = NULL,
  cardinality = c("full", "sample"),
  cascade = NULL,
  detectors = NULL,
  lexicons = default_lexicons(),
  threshold = 0.8,
  seed = 1L
)

Arguments

x

A column (any vector).

name

Optional column header, used by the detector's name tie-break and the header feature block.

n_rows

Row-sample size for the expensive feature blocks and detection. A few hundred is enough for format detection; see the package notes on the speed/coverage trade-off.

blocks

Explicit feature blocks to compute; overrides features_used.

features_used

Character vector of the feature names the fitted model references; blocks are derived from these via fx_feature_manifest(). Ignored when blocks is supplied.

manifest

Optional precomputed fx_feature_manifest() (pass it in a batch so it is not rebuilt per column).

cardinality

"full" (default) recomputes the whole-column cardinality features (n_unique, unique_ratio, hapax_ratio, ...) with one extra linear pass, so those features are exact even though the string blocks were sampled. "sample" skips that pass entirely – fastest, but cardinality features then describe the sample, not the column.

cascade

Optional scorer: a function of the one-row feature data frame returning a list with at least guess, confidence, and ontology. When NULL, falls back to the rule-based guess_column().

detectors

Named list of detector functions; pass the result of data_type_detectors() once in a batch to avoid re-resolving ~90 functions per column. Ignored when cascade is supplied.

lexicons

Lexicon set for the lexicon block; defaults to default_lexicons().

threshold

Minimum detector pass rate to trust a value guess.

seed

Sampling seed (features and detection stay reproducible).

Value

A list: column, source, guess, confidence, ontology, features (one-row data frame), blocks, n (usable rows seen for cardinality), n_unique, and n_sampled.

See also

fx_classify_df(), fx_extract_features(), guess_column()