Class ExpressionsBasedModel.Integration<S extends Optimisation.Solver>

java.lang.Object

org.ojalgo.optimisation.ExpressionsBasedModel.Integration<S>

All Implemented Interfaces:

Optimisation, Optimisation.Integration<ExpressionsBasedModel, S>

Direct Known Subclasses:

ConvexSolver.ModelIntegration, GomorySolver.ModelIntegration, IntegerSolver.ModelIntegration, LinearSolver.ModelIntegration, ServiceIntegration

Enclosing class:

ExpressionsBasedModel

public abstract static class ExpressionsBasedModel.Integration<S extends Optimisation.Solver>
extends Object
implements Optimisation.Integration<ExpressionsBasedModel, S>

Simplifies creating Optimisation.Integrations by providing some default implementations and helper methods for common cases.

The defaults assume that the solver state and model state (variable indices) are identical.

The helper methods are for the very common case when the solver only works with free (not eliminated by the pre-solver) variables, and need to map between the solver state and the full model state.

Author:

apete

Nested Class Summary

Nested classes/interfaces inherited from interface Optimisation

Optimisation.Constraint, Optimisation.ConstraintType, Optimisation.Environment, Optimisation.Integration<M,S>, Optimisation.Model, Optimisation.Objective, Optimisation.Options, Optimisation.ProblemStructure, Optimisation.Result, Optimisation.Sense, Optimisation.Solver, Optimisation.State

Field Summary

Fields inherited from interface Optimisation

ENVIRONMENT

Constructor Summary

Constructors

Constructor	Description
Integration()

Method Summary

Modifier and Type	Method	Description
protected static double	computeReducedCostFromMultipliers(ExpressionsBasedModel model, int variableIndex, Optimisation.Result solverState)	Reconstructs a model-level reduced cost from first principles: rc_v = c_v - Σ a_iv · λ_i, using the variable's objective coefficient and the constraint multipliers reported on solverState.
protected static Optimisation.Result	expandFreeToFull(Optimisation.Result solverState, ExpressionsBasedModel model, DenseArray.Factory<?,?> factory)
protected static Optimisation.Result	expandFreeToFull(Optimisation.Result solverState, ExpressionsBasedModel model, DenseArray.Factory<?,?> factory, Optional<Supplier<Access1D<?>>> reducedGradient)
protected static Optimisation.Result	expandFreeToFull(Optimisation.Result solverState, ExpressionsBasedModel model, DenseArray.Factory<?,?> factory, Optional<Supplier<Access1D<?>>> reducedGradient, Optimisation.Sense solverSense)
final Optimisation.Result	extractSolverState(ExpressionsBasedModel model)	Deprecated. v57 Not needed.
protected int	getIndexInSolver(ExpressionsBasedModel model, Variable variable)	This implementation returns the index of the variable in the model.
protected static int	getIndexOfFreeInSolver(ExpressionsBasedModel model, Variable variable)
protected static double	getObjectiveAdjustment(ExpressionsBasedModel model)	Reads the invalid reference stashed model-sense objective offset.
protected Optimisation.Sense	getSolverSense()	Some solvers are hard-wired to solve either min or max problems.
protected static final boolean	isSwitch(ExpressionsBasedModel model, ExpressionsBasedModel.IntegrationProperty property)
Optimisation.Result	prepareSolverCandidate(Optimisation.Result candidateModelState, ExpressionsBasedModel model)	Preserves the historical kick-starter behaviour: when no candidate is supplied one is derived (cheaply) from the model, then converted to solver state.
protected static Optimisation.Result	reduceFullToFree(Optimisation.Result modelState, ExpressionsBasedModel model, DenseArray.Factory<?,?> factory)
protected static final void	setSwitch(ExpressionsBasedModel model, ExpressionsBasedModel.IntegrationProperty property, boolean value)
Optimisation.Result	toModelState(Optimisation.Result solverState, ExpressionsBasedModel model)	The reverse of toSolverState(Optimisation.Result, ExpressionsBasedModel).
Optimisation.Result	toSolverState(Optimisation.Result modelState, ExpressionsBasedModel model)	This default implementation assumes the solver state and model state are identical, and simply returns the supplied model state.
final ExpressionsBasedModel.Integration<S>	withCapabilityPredicate(Predicate<ExpressionsBasedModel> capabilityPredicate)	Use this to limit the cases where this Optimisation.Integration would be used.
final ExpressionsBasedModel.Integration<S>	withOptionsModifier(Consumer<Optimisation.Options> optionsModifier)	Intercept and modify the Optimisation.Options instance before building the solver.

Methods inherited from class Object

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

Methods inherited from interface Optimisation.Integration

build, isCapable

Constructor Details

Integration

public Integration()

Method Details

computeReducedCostFromMultipliers

protected static double computeReducedCostFromMultipliers(ExpressionsBasedModel model,
 int variableIndex,
 Optimisation.Result solverState)

Reconstructs a model-level reduced cost from first principles: rc_v = c_v - Σ a_iv · λ_i, using the variable's objective coefficient and the constraint multipliers reported on solverState. Useful for variables that the solver doesn't see (eliminated by presolve), and for solver paths whose internal index space prevents direct pass-through of the rc.

Returned value is in the same sense the multipliers are expressed in (i.e. typically MIN, the solver's internal sense). Callers should negate for MAX models if appropriate.

expandFreeToFull

protected static Optimisation.Result expandFreeToFull(Optimisation.Result solverState,
 ExpressionsBasedModel model,
 DenseArray.Factory<?,?> factory)

expandFreeToFull

protected static Optimisation.Result expandFreeToFull(Optimisation.Result solverState,
 ExpressionsBasedModel model,
 DenseArray.Factory<?,?> factory,
 Optional<Supplier<Access1D<?>>> reducedGradient)

expandFreeToFull

protected static Optimisation.Result expandFreeToFull(Optimisation.Result solverState,
 ExpressionsBasedModel model,
 DenseArray.Factory<?,?> factory,
 Optional<Supplier<Access1D<?>>> reducedGradient,
 Optimisation.Sense solverSense)

Parameters:

solverSense - the Optimisation.Sense the solver internally optimises in (usually fixed per solver — most simplex/QP solvers minimise). If this differs from the model's optimisation sense, the reduced gradient values are negated when mapped back to the model so that callers always see them in the model's sense.

getIndexOfFreeInSolver

protected static int getIndexOfFreeInSolver(ExpressionsBasedModel model,
 Variable variable)

getObjectiveAdjustment

protected static double getObjectiveAdjustment(ExpressionsBasedModel model)

Reads the

invalid reference

stashed

model-sense objective offset. Returns 0 if nothing has been stashed.

isSwitch

protected static final boolean isSwitch(ExpressionsBasedModel model,
 ExpressionsBasedModel.IntegrationProperty property)

reduceFullToFree

protected static Optimisation.Result reduceFullToFree(Optimisation.Result modelState,
 ExpressionsBasedModel model,
 DenseArray.Factory<?,?> factory)

setSwitch

protected static final void setSwitch(ExpressionsBasedModel model,
 ExpressionsBasedModel.IntegrationProperty property,
 boolean value)

extractSolverState

@Deprecated
public final Optimisation.Result extractSolverState(ExpressionsBasedModel model)

Deprecated.

v57 Not needed. Use prepareSolverCandidate(org.ojalgo.optimisation.Optimisation.Result, ExpressionsBasedModel) instead

Description copied from interface: Optimisation.Integration

Extract state from the model and convert it to solver state.

Specified by:

extractSolverState in interface Optimisation.Integration<ExpressionsBasedModel, S extends Optimisation.Solver>

prepareSolverCandidate

public Optimisation.Result prepareSolverCandidate(Optimisation.Result candidateModelState,
 ExpressionsBasedModel model)

Preserves the historical kick-starter behaviour: when no candidate is supplied one is derived (cheaply) from the model, then converted to solver state. Solvers that ignore the kick-starter should override this to always return null.

Specified by:

prepareSolverCandidate in interface Optimisation.Integration<ExpressionsBasedModel, S extends Optimisation.Solver>

toModelState

public Optimisation.Result toModelState(Optimisation.Result solverState,
 ExpressionsBasedModel model)

The reverse of toSolverState(Optimisation.Result, ExpressionsBasedModel).

Specified by:

toModelState in interface Optimisation.Integration<ExpressionsBasedModel, S extends Optimisation.Solver>

See Also:

• reduceFullToFree(Optimisation.Result, ExpressionsBasedModel, DenseArray.Factory)
• expandFreeToFull(Optimisation.Result, ExpressionsBasedModel, DenseArray.Factory, Optional)

toSolverState

public Optimisation.Result toSolverState(Optimisation.Result modelState,
 ExpressionsBasedModel model)

This default implementation assumes the solver state and model state are identical, and simply returns the supplied model state.

In any case where the set of variables present in the solver does not match what's in the model one-to-one, this method and its reciprocal toModelState(Optimisation.Result, ExpressionsBasedModel) needs to be overridden with custom mapping implementations.

A very common case is when the solver only works with free (not eliminated by the pre-solver) variables. There are helper methods to do just that.

Specified by:

toSolverState in interface Optimisation.Integration<ExpressionsBasedModel, S extends Optimisation.Solver>

See Also:

• reduceFullToFree(Optimisation.Result, ExpressionsBasedModel, DenseArray.Factory)
• expandFreeToFull(Optimisation.Result, ExpressionsBasedModel, DenseArray.Factory, Optional)

withCapabilityPredicate

public final ExpressionsBasedModel.Integration<S> withCapabilityPredicate(Predicate<ExpressionsBasedModel> capabilityPredicate)

Use this to limit the cases where this Optimisation.Integration would be used.

Returns a new Integration instance where the supplied Predicate needs to test true in addition to the underlying Optimisation.Integration.isCapable(Optimisation.Model).

withOptionsModifier

public final ExpressionsBasedModel.Integration<S> withOptionsModifier(Consumer<Optimisation.Options> optionsModifier)

Intercept and modify the Optimisation.Options instance before building the solver.

getIndexInSolver

protected int getIndexInSolver(ExpressionsBasedModel model,
 Variable variable)

This implementation returns the index of the variable in the model. Override if you need something else. The utility getIndexOfFreeInSolver(ExpressionsBasedModel, Variable) can be used if the solver only works with free variables.

Returns:

The index with which one can reference parameters related to this variable in the solver.

getSolverSense

protected Optimisation.Sense getSolverSense()

Some solvers are hard-wired to solve either min or max problems. (Typically ojAlgo's built-in solvers are implemented that way.) If that's the case this method should indicate what the convention is. Returning null indicates there is no convention/hard-wire, and the solver will either min or max depending what's specified.

When the solver's sense is hard-wired some things, like the sign of the optimal value or the reduced costs, may have to be adjusted.

Since this was a late addition to this class a default implementation was needed, and null is commonly the correct return value for 3:d party solvers.