Crate crucible 

crucible-models

Trained-model packs and the training pipeline for the Converge Engine.

Unlike prism-analytics (closed-form, hand-authored inference), Crucible models are learned from data using the training backend that fits the model family: Burn for differentiable packs and linfa for classical models such as trees and forests. Parameters are stored as trained artifacts, not hand-authored rules.

Training pipeline

The training pipeline lives in training and the supporting data plumbing lives in ingest. The pipeline is composed of Suggestor-shaped agents (DatasetAgent, DataValidationAgent, FeatureEngineeringAgent, HyperparameterSearchAgent, ModelTrainingAgent, ModelEvaluationAgent, ModelRegistryAgent, MonitoringAgent, DeploymentAgent, SampleInferenceAgent) that today run from a binary entrypoint and can be lifted into a Formation when a real retrain trigger pulls.

Planned packs

• trees — Decision Tree classifier (CART, Gini / information gain)
• ensembles — Random Forest and gradient-boosted trees (XGBoost-style)
• svm — Support Vector Machine with kernel functions
• neuro_fuzzy — ANFIS (Adaptive Neuro-Fuzzy Inference System) via Burn

Re-exports

pub use ensembles::RandomForestConfig;

pub use ensembles::RandomForestModel;

pub use model::ClassifierModel;

pub use provenance::CRUCIBLE_PROVENANCE;

pub use provenance::Crucible;

pub use suggestor::ClassifierSuggestor;

pub use training::DataValidationAgent;

pub use training::DatasetAgent;

pub use training::DeploymentAgent;

pub use training::FeatureEngineeringAgent;

pub use training::HyperparameterSearchAgent;

pub use training::ModelEvaluationAgent;

pub use training::ModelRegistryAgent;

pub use training::ModelTrainingAgent;

pub use training::MonitoringAgent;

pub use training::SampleInferenceAgent;

pub use trees::DecisionTreeClassifier;

pub use trees::DecisionTreeConfig;

pub use types::ClassPredictionPayload;

pub use types::ClassificationFeaturesPayload;

Modules

ensembles

Ensemble models: Random Forest and gradient-boosted trees.

fixtures

Deterministic synthetic fixtures for crucible-models.

ingest

File ingest: read CSV, Parquet, and Excel into Polars DataFrames.

model

Model trait for trained-artifact packs in crucible-models.

neuro_fuzzy

ANFIS — Adaptive Neuro-Fuzzy Inference System.

provenance

Crucible’s ProvenanceSource marker.

suggestor

Crucible classifier Suggestor: bridges a trained ClassifierModel into the Convergence Engine via the typed payload contract.

svm

Support Vector Machine (SVM) with kernel functions.

training

trees

Decision Tree classifier (CART — Classification and Regression Trees).

types

Crucible-local typed payloads for ProposedFact content.

Type Aliases

DecisionTreeClassifierSuggestor

Inference Suggestor for the DecisionTreeClassifier pack.

RandomForestClassifierSuggestor

Inference Suggestor for the RandomForestModel pack.