Extract the value-profile feature vector for a column

Profiles a column into a wide, flat numeric feature row across a set of independent, block-gated feature families (core, sequence, length, character-class, punctuation, mask, character-frequency, regex, numeric, token, lexicon, header). This is the substrate the learned router and the signature primitives are built on. Extraction is reproducible (a local, non-invasive RNG stream) so a profile is stable across runs.

Usage

fx_extract_features(
  x,
  col_name = NULL,
  blocks = ALL_BLOCKS,
  lexicons = default_lexicons(),
  max_unique = 20000,
  seed = 1L
)

Arguments

x

A column (coerced to character).

col_name

Optional column name; enables the header block features.

blocks

Character vector of feature blocks to compute (default all). Compute only the blocks a fitted model references to trim deployment cost (see fx_feature_manifest(), fx_blocks_for_features()).

lexicons

Named list of controlled vocabularies for the lexicon block (default default_lexicons()).

max_unique

Cap on distinct values profiled (downsampled above it).

seed

Integer seed for the internal, caller-non-invasive RNG stream.

Value

A one-row data frame of numeric/NA features.

See also

fx_type_signature(), fx_feature_manifest()