ForestFire core model, training, inference, and interchange layer.
The crate is organized around a few stable abstractions:
- [
forestfire_data::TableAccess] is the common data boundary for both training and inference. - [
TrainConfig] is the normalized configuration surface shared by the Rust and Python APIs. - [
Model] is the semantic model view used for exact prediction, serialization, and introspection. - [
OptimizedModel] is a lowered runtime view used when prediction speed matters more than preserving the original tree layout.
Keeping the semantic model and the runtime model separate is deliberate. It makes export and introspection straightforward while still allowing the optimized path to use layouts that are awkward to serialize directly.