Module modular_framework 

Modular Framework for Linear Models

This module implements a trait-based system for pluggable solvers, loss functions, and regularization schemes. This addresses TODO items for architectural improvements:

• Separate solver implementations into trait-based system
• Create pluggable loss function framework
• Implement composable regularization schemes
• Add extensible prediction interface

Structs

BayesianPredictionProvider

Bayesian prediction provider with uncertainty quantification

CompositeObjective

A composite objective that combines a loss function with optional regularization

LinearPredictionProvider

Standard linear prediction provider

ModularConfig

Configuration for the modular framework

ModularFramework

The main modular framework that coordinates components

ModularLinearModel

A linear model built using the modular framework

ObjectiveData

Data structure containing all information needed for objective computation

ObjectiveMetadata

Metadata for objective computation

OptimizationResult

Result of optimization through the modular framework

PredictionWithConfidence

Prediction result with confidence intervals

PredictionWithUncertainty

Prediction result with uncertainty quantification

ProbabilisticPredictionProvider

Probabilistic prediction provider for classification

SolverInfo

Information about the solver execution

SolverRecommendations

Recommendations for solver configuration

Traits

LossFunction

Trait for loss functions that measure prediction error

Objective

Trait for optimization objectives that can be minimized

OptimizationSolver

Trait for optimization solvers

PredictionProvider

Extensible prediction interface supporting different prediction types

Regularization

Trait for regularization penalties

Functions

create_modular_linear_regression

Utility function to create a modular linear regression solver