discopt.llm.reformulation

Auto-reformulation engine.

Feature 3C: LLM identifies WHICH reformulations to apply; deterministic code executes them (correctness guaranteed).

Supported reformulations: - Big-M tightening - McCormick substitution for bilinear terms - Symmetry breaking constraints - Variable bound tightening from constraint structure

Attributes

logger

Classes

ReformulationSuggestion

A suggested reformulation with before/after description.

Functions

analyze_reformulations(→ list[ReformulationSuggestion])	Analyze a model and suggest reformulations.
apply_bound_tightening(→ int)	Apply deterministic bound tightening to the model.

Module Contents

discopt.llm.reformulation.logger

class discopt.llm.reformulation.ReformulationSuggestion

A suggested reformulation with before/after description.

category: str

description: str

impact: str

auto_applicable: bool = False

details: dict

discopt.llm.reformulation.analyze_reformulations(model: discopt.modeling.core.Model, llm: bool = False, llm_model: str | None = None) → list[ReformulationSuggestion]

Analyze a model and suggest reformulations.

Combines deterministic analysis (always runs) with optional LLM-powered analysis (when llm=True).

Parameters:

• model (Model) – The model to analyze.

• llm (bool, default False) – Enable LLM-powered analysis.

• llm_model (str, optional) – LLM model string.

Returns:

Ordered by expected impact (highest first).

Return type:

list of ReformulationSuggestion

discopt.llm.reformulation.apply_bound_tightening(model: discopt.modeling.core.Model) → int

Apply deterministic bound tightening to the model.

Derives tighter variable bounds from constraint structure. This is safe — it only tightens bounds, never loosens them.

Parameters:

model (Model) – The model to tighten (modified in place).

Returns:

Number of bounds tightened.

Return type:

int