Expand description
Lean, pure-Rust classical machine learning.
FerricML focuses first on linear models and random forests, with stable estimator semantics and fast, allocation-conscious inference.
Modulesยง
- api
- Shared estimator contracts and model errors.
- artifact
- Stable, bounded model-artifact errors and format identity.
- calibration
- Post-hoc probability calibration around an already-fitted classifier.
- data
- Validated data containers, and the shared vocabulary over them.
- dummy
- Baseline estimators that ignore their features.
- ensemble
- Ensemble estimators.
- inspection
- Model-agnostic inspection built on the prediction and scoring contracts.
- linear_
model - Linear estimators with stable fit and prediction semantics.
- metrics
- Deterministic classification and regression metrics.
- model_
selection - Splitting, scoring, cross-validation, and hyperparameter search.
- pipeline
- Generic fitted preprocessing pipelines.
- preprocessing
- Fitted dense preprocessing transformers.
- ranking
- Pairwise linear ranking and tie-aware rank metrics.
- tree
- Decision trees, and the private grower every tree-shaped estimator shares.