Package jline.opt.solver

Class LineOptSolverOptions

java.lang.Object

jline.opt.solver.LineOptSolverOptions

public class LineOptSolverOptions
extends Object

Configuration options for LineOptSolver, mirroring the option dict of native-Python LineOptSolver.defaultOptions(). Chained setters return this.

Field Summary

Fields

Modifier and Type	Field	Description
int	fdRefresh	partial_plus_fd full finite-difference correction period.
double	fdStep
List<String>	frozenLayers	Explicit layer freezing (LQN only): host/task layer names whose variables are held at the model's current value instead of being optimized.
int	gradientRestarts
String	lqnGradient	LayeredNetwork (LQN) gradient source, used only when the model is a LayeredNetwork and the gradient path is taken: 'fd' (whole-model finite difference, robust default), 'partial_sens' (SolverLN per-layer partial derivatives, cheap/biased), or 'partial_plus_fd' (partial with a periodic full-FD correction every fdRefresh gradient evaluations).
int	maxIterations
double	mutationHigh
double	mutationLow
String	optimizer	'evolution', 'gradient', or 'auto'.
double	penaltyWeight
int	popsize
double	recombination
String	scenarioAggregation	'worst' (minimax) or 'mean' (weighted) aggregation across scenarios.
Long	seed
String	strategy
double	timeLimit
double	tol
boolean	verbose

Constructor Summary

Constructors

Constructor	Description
LineOptSolverOptions()

Method Summary

Modifier and Type	Method	Description
LineOptSolverOptions	setFdRefresh(int v)	Set the partial_plus_fd full finite-difference correction period.
LineOptSolverOptions	setFdStep(double v)
LineOptSolverOptions	setFrozenLayers(List<String> layers)	Freeze the given LQN layers (list of host/task layer names).
LineOptSolverOptions	setGradientRestarts(int v)
LineOptSolverOptions	setLqnGradient(String v)	Set the LQN gradient source: 'fd', 'partial_sens', or 'partial_plus_fd'.
LineOptSolverOptions	setMaxIterations(int v)
LineOptSolverOptions	setMutation(double low, double high)
LineOptSolverOptions	setOptimizer(String v)
LineOptSolverOptions	setPenaltyWeight(double v)
LineOptSolverOptions	setPopsize(int v)
LineOptSolverOptions	setRecombination(double v)
LineOptSolverOptions	setScenarioAggregation(String v)
LineOptSolverOptions	setSeed(long v)
LineOptSolverOptions	setStrategy(String v)
LineOptSolverOptions	setTimeLimit(double v)
LineOptSolverOptions	setTolerance(double v)
LineOptSolverOptions	setVerbose(boolean v)

Methods inherited from class java.lang.Object

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

Field Details

strategy

public String strategy

popsize

public int popsize

mutationLow

public double mutationLow

mutationHigh

public double mutationHigh

recombination

public double recombination

tol

public double tol

maxIterations

public int maxIterations

timeLimit

public double timeLimit

seed

public Long seed

verbose

public boolean verbose

penaltyWeight

public double penaltyWeight

scenarioAggregation

public String scenarioAggregation

'worst' (minimax) or 'mean' (weighted) aggregation across scenarios.

optimizer

public String optimizer

'evolution', 'gradient', or 'auto'.

fdStep

public double fdStep

gradientRestarts

public int gradientRestarts

lqnGradient

public String lqnGradient

LayeredNetwork (LQN) gradient source, used only when the model is a LayeredNetwork and the gradient path is taken: 'fd' (whole-model finite difference, robust default), 'partial_sens' (SolverLN per-layer partial derivatives, cheap/biased), or 'partial_plus_fd' (partial with a periodic full-FD correction every fdRefresh gradient evaluations).

fdRefresh

public int fdRefresh

partial_plus_fd full finite-difference correction period.

frozenLayers

public List<String> frozenLayers

Explicit layer freezing (LQN only): host/task layer names whose variables are held at the model's current value instead of being optimized. Null (or empty) freezes nothing.

Constructor Details

LineOptSolverOptions

public LineOptSolverOptions()

Method Details

setStrategy

public LineOptSolverOptions setStrategy(String v)

setPopsize

public LineOptSolverOptions setPopsize(int v)

setMutation

public LineOptSolverOptions setMutation(double low,
 double high)

setRecombination

public LineOptSolverOptions setRecombination(double v)

setTolerance

public LineOptSolverOptions setTolerance(double v)

setMaxIterations

public LineOptSolverOptions setMaxIterations(int v)

setTimeLimit

public LineOptSolverOptions setTimeLimit(double v)

setSeed

public LineOptSolverOptions setSeed(long v)

setVerbose

public LineOptSolverOptions setVerbose(boolean v)

setPenaltyWeight

public LineOptSolverOptions setPenaltyWeight(double v)

setScenarioAggregation

public LineOptSolverOptions setScenarioAggregation(String v)

setOptimizer

public LineOptSolverOptions setOptimizer(String v)

setFdStep

public LineOptSolverOptions setFdStep(double v)

setGradientRestarts

public LineOptSolverOptions setGradientRestarts(int v)

setLqnGradient

public LineOptSolverOptions setLqnGradient(String v)

Set the LQN gradient source: 'fd', 'partial_sens', or 'partial_plus_fd'.

setFdRefresh

public LineOptSolverOptions setFdRefresh(int v)

Set the partial_plus_fd full finite-difference correction period.

setFrozenLayers

public LineOptSolverOptions setFrozenLayers(List<String> layers)

Freeze the given LQN layers (list of host/task layer names).