Package jline.opt.decomposition

Class DecompositionWorkflow

java.lang.Object

jline.opt.decomposition.DecompositionWorkflow

public class DecompositionWorkflow
extends Object

Decomposes a joint optimization problem into per-variable-type subproblems solved via Gauss-Seidel cycling with fixed-value propagation. An internal topological sort (no external dependency) orders subproblems when explicit dependencies are set. Mirrors native-Python line_solver.opt.decomposition.DecompositionWorkflow.

Constructor Summary

Constructors

Constructor	Description
DecompositionWorkflow(OptimizationProblem problem)

Method Summary

Modifier and Type	Method	Description
DecompositionWorkflow	addSubProblem(String name, List<DecisionVariable> variables, List<String> after)
DecompositionWorkflow	autoDecompose()
OptimizationProblem	getProblem()
List<SubProblem>	getSubProblems()
DecompositionWorkflow	setDependency(String fromProblem, String toProblem)
DecompositionWorkflow	setSolverOptions(LineOptSolverOptions options)
WorkflowResult	solveHierarchical()
WorkflowResult	solveLayered()	LQN layer-wise decomposition with default settings.
WorkflowResult	solveLayered(int maxCycles, double tolerance, boolean autoFreeze, double freezeTol, List<String> frozenLayersSeed)	Solve an LQN by layer, optionally freezing converged layers.
WorkflowResult	solveSequential(int maxCycles, double tolerance)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

DecompositionWorkflow

public DecompositionWorkflow(OptimizationProblem problem)

Method Details

getProblem

public OptimizationProblem getProblem()

getSubProblems

public List<SubProblem> getSubProblems()

setSolverOptions

public DecompositionWorkflow setSolverOptions(LineOptSolverOptions options)

autoDecompose

public DecompositionWorkflow autoDecompose()

setDependency

public DecompositionWorkflow setDependency(String fromProblem,
 String toProblem)

addSubProblem

public DecompositionWorkflow addSubProblem(String name,
 List<DecisionVariable> variables,
 List<String> after)

solveSequential

public WorkflowResult solveSequential(int maxCycles,
 double tolerance)

solveHierarchical

public WorkflowResult solveHierarchical()

solveLayered

public WorkflowResult solveLayered()

LQN layer-wise decomposition with default settings.

solveLayered

public WorkflowResult solveLayered(int maxCycles,
 double tolerance,
 boolean autoFreeze,
 double freezeTol,
 List<String> frozenLayersSeed)

Solve an LQN by layer, optionally freezing converged layers. Groups decision variables by the LQN layer they perturb (host or task layer) and cycles Gauss-Seidel over the layer groups, fixing every other layer's variables at their current values while one layer is optimized. This is the LQN analogue of solveSequential(int, double), but the subproblems are LAYERS.

Freezing has two composable sources: frozenLayersSeed, an explicit set held fixed throughout; and autoFreeze, adaptive -- after each cycle a layer whose representative node metrics moved less than freezeTol (relative) is frozen and skipped, and unfrozen again if any still-active layer later moves by more than freezeTol.

Convergence is on the full penalized objective delta between cycles, or when every layer is frozen. Falls back to solveSequential(int, double) for a flat network. The returned WorkflowResult carries the extra fields frozenLayers (final frozen set) and modelEvaluations (total LINE solves). Mirrors native-Python solveLayered.

Parameters:

maxCycles - maximum Gauss-Seidel cycles

tolerance - objective-delta convergence tolerance

autoFreeze - enable adaptive layer freezing

freezeTol - relative-movement threshold for freezing a layer

frozenLayersSeed - explicit layers held fixed throughout (may be null)

Returns:

the workflow result with LQN freezing diagnostics