Expand description
xlog-induce — native bounded exact-induction engine.
Scores all (left, right) candidate pairs for the four supported topologies (chain,
star, fanout, fanin) in a single batched GPU pass and returns the top-K per topology
with full candidate metadata.
Behaviorally equivalent to the backend="python" reference implementation
in crates/pyxlog/python/pyxlog/ilp/exact_induce.py on bounded requests;
the parity contract is locked by python/tests/test_ilp_exact_induce.py.
The native production path includes request validation, deterministic reduction, trivial-dead-end early returns, the batched scoring kernel, device-side top-K selection, and compact selected-row transfers.
Re-exports§
pub use provenance::InductionProvenanceRegistry;pub use reduce::reduce_per_topology;pub use reduce::ScoredPair;pub use types::ExactInductionConfig;pub use types::ExactInductionResult;pub use types::InducedRuleProvenance;pub use types::InducedRuleRegistry;pub use types::InductionAlternative;pub use types::InductionSupportRow;pub use types::RuleSourceKind;pub use types::ScoredCandidate;pub use types::Topology;
Modules§
- index
- Request-local device index construction from compiled relation buffers.
- provenance
- Compatibility aliases for generated induction provenance.
- reduce
- Deterministic per-topology top-K reduction and tie diagnostics.
- score
- Batched topology scoring orchestration over request-local indexes.
- types
- Public types for the exact-induction engine.
Structs§
- Induce
Exact Request - Inputs to one
induce_exactcall.
Functions§
- induce_
exact - Run exact induction against one request.