Crate ferrolearn_core 

ferrolearn-core

Core traits, error types, dataset abstractions, and pipeline infrastructure for the ferrolearn machine learning framework.

This crate defines the foundational abstractions that all other ferrolearn crates depend on:

• Fit, Predict, Transform, FitTransform — the core ML traits with compile-time enforcement that predict() cannot be called on an unfitted model.
• FerroError — the unified error type with rich diagnostic context.
• Dataset — a trait for querying tabular data shape, with implementations for ndarray::Array2<f32> and ndarray::Array2<f64>.
• pipeline::Pipeline — a dynamic-dispatch pipeline that composes transformers and a final estimator.
• Introspection traits — HasCoefficients, HasFeatureImportances, HasClasses for inspecting fitted model internals.

Design Principles

Compile-Time Safety (AC-3)

The unfitted configuration struct (e.g., LinearRegression) implements Fit but not Predict. Calling fit() returns a new fitted type (e.g., FittedLinearRegression) that implements Predict. Attempting to call predict() on an unfitted model is a compile error.

Float Generics (REQ-15)

All algorithms are generic over F: num_traits::Float + Send + Sync + 'static.

Error Handling

All public functions return Result<T, FerroError>. Library code never panics.

Pluggable Backends (REQ-19)

The Backend trait abstracts linear algebra operations (SVD, QR, Cholesky, etc.), allowing algorithms to be generic over the backend implementation. The default backend NdarrayFaerBackend delegates to the faer crate.

Re-exports

pub use backend::Backend;

pub use backend_faer::NdarrayFaerBackend;

pub use dataset::Dataset;

pub use error::FerroError;

pub use error::FerroResult;

pub use introspection::HasClasses;

pub use introspection::HasCoefficients;

pub use introspection::HasFeatureImportances;

pub use traits::Fit;

pub use traits::FitTransform;

pub use traits::PartialFit;

pub use traits::Predict;

pub use traits::Transform;

Modules

backend

Pluggable backend trait for linear algebra operations.

backend_faer

Default backend implementation using faer for linear algebra.

dataset

Dataset trait and implementations for common array types.

error

Error types for the ferrolearn framework.

introspection

Introspection traits for fitted models.

pipeline

Dynamic-dispatch pipeline for composing transformers and estimators.

traits

Core ML traits for the ferrolearn framework.

typed_pipeline

Compile-time type-safe pipeline using recursive tuple types.

Type Aliases

DefaultBackend

The default linear algebra backend.