Crate napparent_tabular

Expand description

Tabular effect features on Apache Arrow batches (Barn Effect algorithm).

Discretize columns, accumulate pairwise co-occurrence statistics, and produce per-row {column}_effect attributions. See transform_record_batches and transform_record_batches_chunked.

Structs§

ActivationConfig: Activation settings for both pipeline stages.
BatchChunk: Feature columns from one batch — Arrow buffers shared via ArrayRef where possible.
BinDepth
CancelToken: Checked at the start of each batch iteration; safe to share across threads.
ChunkTable: Feature matrix columns (no target column), in schema order.
ColGraph: Maps column index in ChunkTable -> original name; which indices are dropped from preprocessing.
ColumnPreprocess
CtrlcGuard: Restores a no-op SIGINT handler when dropped.
EffectContext: Context passed to effect activations per row.
PairAggregator
PairStats: Raw accumulated stats for one canonical unordered value-pair key in vals_map.
PreprocessStream
TransformConfig: Full configuration for crate::pipeline::transform_record_batches.
TransformLimits: Optional fail-fast caps for large or wide transforms (all None = no limits).

BatchColumn: Column storage for one batch pass.
BinType
ColumnVec
EffectActivation: How combined column signal becomes an effect feature value.
KgPairActivation: How a KG edge weight is computed from accumulated pair counts.
OutcomeSource: Outcome vector aligned with table rows.
OutcomesRef: Borrowed outcomes for zero-copy aggregation when target is null-free Float32.
TargetColumn: Target column — shared Float32 buffer or owned fallback.
ValueKey

batch_from_map: Build a RecordBatch from string-name → column map (mixed float / utf8).
concat_same_schema
round_to_significant_figures: Match preprocess_stream.round_to_significant_figures / NumPy // 1 on log scale.
split_batch_views: Zero-copy batch split: feature columns share Arrow buffers via ArrayRef.
split_batch_xy: Extract feature table X, target Y, and column metadata (materializes owned columns).
transform_record_batches: Chunked pipeline: preprocess passes, pair aggregation, then apply per batch.
transform_record_batches_chunked: Chunked pipeline: preprocess, pair aggregation, apply — returns one output batch per input batch.