Expand description
Stable intermediate representation for ForestFire models.
The IR sits between two worlds:
- the semantic in-memory model types used for training and introspection
- the lowered runtime structures used by optimized inference
It exists so models can be serialized, schema-checked, inspected from other languages, and reconstructed without depending on the exact Rust memory layout of the training structs.
Structs§
- Aggregation
- Binary
Children - Bool
Encoding - Compatibility
- Determinism
- Indexed
Leaf - Inference
Options - Input
Feature - Input
Schema - Integrity
Section - Leaf
Indexing - Model
Package Ir - Top-level model package serialized by the library.
- Model
Section - Structural model description independent of any concrete runtime layout.
- Multiway
Branch - Multiway
Split - Node
Stats - Numeric
Binning - Oblivious
Level - Output
Field - Output
Schema - Postprocessing
Section - Preprocessing
Section - Producer
Metadata - TieBreaking
- Training
Metadata - Serialized training metadata reflected back to bindings and docs.
Enums§
- Binary
Split - Feature
Binning - IrError
- Leaf
Payload - Node
Tree Node - Oblivious
Split - Postprocessing
Step - Tree
Definition - Concrete tree payload stored in the IR.