Struct FieldComparator 

pub struct FieldComparator { /* private fields */ }

Pairwise field comparator that applies similarity functions to produce a field-major ComparisonBatch.

Implementations

impl FieldComparator

pub fn from_schema(schema: &Schema) -> FieldComparator

pub fn from_mapping( mappings: &[FieldMapping], a_schema: &Schema, ) -> FieldComparator

Build a comparator for cross-schema linkage from an explicit field-mapping list.

Field kinds are inferred from a_schema by looking up each a_field. Fields not found in a_schema default to FieldKind::Categorical.

pub fn compare_pair_mapped( &self, a: &Record, b: &Record, mappings: &[FieldMapping], ) -> ComparisonVector

Compare a cross-schema pair using an explicit field-mapping list.

For each mapping, looks up a_field in record a and b_field in record b. When a field is missing the NullPolicy decides the level: Skip gives Null (EM ignores it), PenaliseAbsence gives None (hard fail).

pub fn compare_batch_mapped( &self, records: &[Record], indices: &[(usize, usize)], mappings: &[FieldMapping], ) -> ComparisonBatch

Batch comparison for cross-schema linkage using explicit field mappings.

Equivalent to calling compare_pair_mapped per pair then assembling the field-major ComparisonBatch. n_fields = mappings.len().

pub fn with_thresholds( self, field_idx: usize, thresholds: LevelThresholds, ) -> FieldComparator

pub fn with_fns( self, field_idx: usize, fns: Vec<Box<dyn SimilarityFn>>, ) -> FieldComparator

pub fn compare_batch_from_pool( &self, pool: &RecordPool, indices: &[(usize, usize)], schema: &Schema, ) -> ComparisonBatch

Pool-native batch comparison, the primary hot path.

Reads RecordPool columns directly: zero HashMap lookups, no Record::clone(). Uses Rayon for parallel per-pair comparison into a flat pair-major buffer (zero per-pair heap allocations), then transposes to the field-major ComparisonBatch layout required by all GPU EM kernels (CUDA/Vulkan/AVX2): levels[f * n_pairs + p].

Trait Implementations

impl Comparator for FieldComparator

fn compare(&self, a: &Record, b: &Record, schema: &Schema) -> ComparisonVector

Compare a single pair, always CPU, returns an individual vector.

fn compare_batch_from_pool( &self, pool: &RecordPool, indices: &[(usize, usize)], schema: &Schema, ) -> ComparisonBatch

Pool-native batch comparison, the primary hot path.