Interface GlopParametersOrBuilder
- All Superinterfaces:
com.google.protobuf.MessageLiteOrBuilder
,com.google.protobuf.MessageOrBuilder
- All Known Implementing Classes:
GlopParameters
,GlopParameters.Builder
@Generated
public interface GlopParametersOrBuilder
extends com.google.protobuf.MessageOrBuilder
-
Method Summary
Modifier and TypeMethodDescriptionboolean
During incremental solve, let the solver decide if it use the primal or dual simplex algorithm depending on the current solution and on the new problem.int
Number of iterations between two basis refactorizations.boolean
If true, the internal API will change the return status to imprecise if the solution does not respect the internal tolerances.optional .operations_research.glop.GlopParameters.CostScalingAlgorithm cost_scaling = 60 [default = CONTAIN_ONE_COST_SCALING];
double
If the starting basis contains FREE variable with bounds, we will move any such variable to their closer bounds if the distance is smaller than this parameter.double
During a degenerate iteration, the more conservative approach is to do a step of length zero (while shifting the bound of the leaving variable).int
Devex weights will be reset to 1.0 after that number of updates.double
Value in the input LP lower than this will be ignored.double
In order to increase the sparsity of the manipulated vectors, floating point values with a magnitude smaller than this parameter are set to zero (only in some places).double
Variables whose reduced costs have an absolute value smaller than this tolerance are not considered as entering candidates.double
When solve_dual_problem is LET_SOLVER_DECIDE, take the dual if the number of constraints of the problem is more than this threshold times the number of variables.boolean
On some problem like stp3d or pds-100 this makes a huge difference in speed and number of iterations of the dual simplex.double
Like small_pivot_threshold but for the dual simplex.boolean
If this is true, then basis_refactorization_period becomes a lower bound on the number of iterations between two refactorization (provided there is no numerical accuracy issues).boolean
Whether or not we exploit the singleton columns already present in the problem when we create the initial basis.PricingRule to use during the feasibility phase.double
This impacts the ratio test and indicates by how much we allow a basic variable value that we move to go out of bounds.What heuristic is used to try to replace the fixed slack columns in the initial basis of the primal simplex.double
If our upper bound on the condition number of the initial basis (from our heurisitic or a warm start) is above this threshold, we revert to an all slack basis.boolean
Whether we initialize devex weights to 1.0 or to the norms of the matrix columns.boolean
If true, logs the progress of a solve to LOG(INFO).boolean
If true, logs will be displayed to stdout instead of using Google log info.double
Threshold for LU-factorization: for stability reasons, the magnitude of the chosen pivot at a given step is guaranteed to be greater than this threshold times the maximum magnitude of all the possible pivot choices in the same column.double
If a pivot magnitude is smaller than this during the Markowitz LU factorization, then the matrix is assumed to be singular.int
How many columns do we look at in the Markowitz pivoting rule to find a good pivot.double
Maximum deterministic time allowed to solve a problem.long
Maximum number of simplex iterations to solve a problem.double
When the solution of phase II is imprecise, we re-run the phase II with the opposite algorithm from that imprecise solution (i.e., if primal or dual simplex was used, we use dual or primal simplex, respectively).double
Maximum time allowed in seconds to solve a problem.double
Any finite values in the input LP must be below this threshold, otherwise the model will be reported invalid.double
We never follow a basis change with a pivot under this threshold.int
Number of threads in the OMP parallel sections.double
The solver will stop as soon as it has proven that the objective is smaller than objective_lower_limit or greater than objective_upper_limit.double
optional double objective_upper_limit = 41 [default = inf];
PricingRule to use during the optimization phase.boolean
When this is true, then the costs are randomly perturbed before the dual simplex is even started.double
A floating point tolerance used by the preprocessors.double
This tolerance indicates by how much we allow the variable values to go out of bounds and still consider the current solution primal-feasible.boolean
If true, then when the solver returns a solution with an OPTIMAL status, we can guarantee that: - The primal variable are in their boundsboolean
If the optimization phases finishes with super-basic variables (i.e., variables that either 1) have bounds but are FREE in the basis, or 2) have no bounds and are FREE in the basis at a nonzero value), then run a "push" phase to push these variables to bounds, obtaining a vertex solution.int
At the beginning of each solve, the random number generator used in some part of the solver is reinitialized to this seed.double
During the primal simplex (resp. dual simplex), the coefficients of the direction (resp. update row) with a magnitude lower than this threshold are not considered during the ratio test.double
Note that the threshold is a relative error on the actual norm (not the squared one) and that edge norms are always greater than 1.double
We estimate the accuracy of the iteratively computed reduced costs.double
We estimate the factorization accuracy of B during each pivot by using the fact that we can compute the pivot coefficient in two ways: - From direction[leaving_row]double
The magnitude of the cost perturbation is given by RandomIn(1.0, 2.0) * ( relative_cost_perturbation * cost + relative_max_cost_perturbation * max_cost);double
optional double relative_max_cost_perturbation = 55 [default = 1e-07];
optional .operations_research.glop.GlopParameters.ScalingAlgorithm scaling_method = 57 [default = EQUILIBRATION];
double
When we choose the leaving variable, we want to avoid small pivot because they are the less precise and may cause numerical instabilities.double
When the problem status is OPTIMAL, we check the optimality using this relative tolerance and change the status to IMPRECISE if an issue is detected.Whether or not we solve the dual of the given problem.boolean
Whether to use absl::BitGen instead of MTRandom.boolean
We have two possible dual phase I algorithms.boolean
Whether or not we use the dual simplex algorithm instead of the primal.boolean
If presolve runs, include the pass that detects implied free variables.boolean
Whether or not to use the middle product form update rather than the standard eta LU update.boolean
Whether or not we use advanced preprocessing techniques.boolean
Whether or not we scale the matrix A so that the maximum coefficient on each line and each column is 1.0.boolean
Whether or not we keep a transposed version of the matrix A to speed-up the pricing at the cost of extra memory and the initial tranposition computation.boolean
During incremental solve, let the solver decide if it use the primal or dual simplex algorithm depending on the current solution and on the new problem.boolean
Number of iterations between two basis refactorizations.boolean
If true, the internal API will change the return status to imprecise if the solution does not respect the internal tolerances.boolean
optional .operations_research.glop.GlopParameters.CostScalingAlgorithm cost_scaling = 60 [default = CONTAIN_ONE_COST_SCALING];
boolean
If the starting basis contains FREE variable with bounds, we will move any such variable to their closer bounds if the distance is smaller than this parameter.boolean
During a degenerate iteration, the more conservative approach is to do a step of length zero (while shifting the bound of the leaving variable).boolean
Devex weights will be reset to 1.0 after that number of updates.boolean
Value in the input LP lower than this will be ignored.boolean
In order to increase the sparsity of the manipulated vectors, floating point values with a magnitude smaller than this parameter are set to zero (only in some places).boolean
Variables whose reduced costs have an absolute value smaller than this tolerance are not considered as entering candidates.boolean
When solve_dual_problem is LET_SOLVER_DECIDE, take the dual if the number of constraints of the problem is more than this threshold times the number of variables.boolean
On some problem like stp3d or pds-100 this makes a huge difference in speed and number of iterations of the dual simplex.boolean
Like small_pivot_threshold but for the dual simplex.boolean
If this is true, then basis_refactorization_period becomes a lower bound on the number of iterations between two refactorization (provided there is no numerical accuracy issues).boolean
Whether or not we exploit the singleton columns already present in the problem when we create the initial basis.boolean
PricingRule to use during the feasibility phase.boolean
This impacts the ratio test and indicates by how much we allow a basic variable value that we move to go out of bounds.boolean
What heuristic is used to try to replace the fixed slack columns in the initial basis of the primal simplex.boolean
If our upper bound on the condition number of the initial basis (from our heurisitic or a warm start) is above this threshold, we revert to an all slack basis.boolean
Whether we initialize devex weights to 1.0 or to the norms of the matrix columns.boolean
If true, logs the progress of a solve to LOG(INFO).boolean
If true, logs will be displayed to stdout instead of using Google log info.boolean
Threshold for LU-factorization: for stability reasons, the magnitude of the chosen pivot at a given step is guaranteed to be greater than this threshold times the maximum magnitude of all the possible pivot choices in the same column.boolean
If a pivot magnitude is smaller than this during the Markowitz LU factorization, then the matrix is assumed to be singular.boolean
How many columns do we look at in the Markowitz pivoting rule to find a good pivot.boolean
Maximum deterministic time allowed to solve a problem.boolean
Maximum number of simplex iterations to solve a problem.boolean
When the solution of phase II is imprecise, we re-run the phase II with the opposite algorithm from that imprecise solution (i.e., if primal or dual simplex was used, we use dual or primal simplex, respectively).boolean
Maximum time allowed in seconds to solve a problem.boolean
Any finite values in the input LP must be below this threshold, otherwise the model will be reported invalid.boolean
We never follow a basis change with a pivot under this threshold.boolean
Number of threads in the OMP parallel sections.boolean
The solver will stop as soon as it has proven that the objective is smaller than objective_lower_limit or greater than objective_upper_limit.boolean
optional double objective_upper_limit = 41 [default = inf];
boolean
PricingRule to use during the optimization phase.boolean
When this is true, then the costs are randomly perturbed before the dual simplex is even started.boolean
A floating point tolerance used by the preprocessors.boolean
This tolerance indicates by how much we allow the variable values to go out of bounds and still consider the current solution primal-feasible.boolean
If true, then when the solver returns a solution with an OPTIMAL status, we can guarantee that: - The primal variable are in their boundsboolean
If the optimization phases finishes with super-basic variables (i.e., variables that either 1) have bounds but are FREE in the basis, or 2) have no bounds and are FREE in the basis at a nonzero value), then run a "push" phase to push these variables to bounds, obtaining a vertex solution.boolean
At the beginning of each solve, the random number generator used in some part of the solver is reinitialized to this seed.boolean
During the primal simplex (resp. dual simplex), the coefficients of the direction (resp. update row) with a magnitude lower than this threshold are not considered during the ratio test.boolean
Note that the threshold is a relative error on the actual norm (not the squared one) and that edge norms are always greater than 1.boolean
We estimate the accuracy of the iteratively computed reduced costs.boolean
We estimate the factorization accuracy of B during each pivot by using the fact that we can compute the pivot coefficient in two ways: - From direction[leaving_row]boolean
The magnitude of the cost perturbation is given by RandomIn(1.0, 2.0) * ( relative_cost_perturbation * cost + relative_max_cost_perturbation * max_cost);boolean
optional double relative_max_cost_perturbation = 55 [default = 1e-07];
boolean
optional .operations_research.glop.GlopParameters.ScalingAlgorithm scaling_method = 57 [default = EQUILIBRATION];
boolean
When we choose the leaving variable, we want to avoid small pivot because they are the less precise and may cause numerical instabilities.boolean
When the problem status is OPTIMAL, we check the optimality using this relative tolerance and change the status to IMPRECISE if an issue is detected.boolean
Whether or not we solve the dual of the given problem.boolean
Whether to use absl::BitGen instead of MTRandom.boolean
We have two possible dual phase I algorithms.boolean
Whether or not we use the dual simplex algorithm instead of the primal.boolean
If presolve runs, include the pass that detects implied free variables.boolean
Whether or not to use the middle product form update rather than the standard eta LU update.boolean
Whether or not we use advanced preprocessing techniques.boolean
Whether or not we scale the matrix A so that the maximum coefficient on each line and each column is 1.0.boolean
Whether or not we keep a transposed version of the matrix A to speed-up the pricing at the cost of extra memory and the initial tranposition computation.Methods inherited from interface com.google.protobuf.MessageLiteOrBuilder
isInitialized
Methods inherited from interface com.google.protobuf.MessageOrBuilder
findInitializationErrors, getAllFields, getDefaultInstanceForType, getDescriptorForType, getField, getInitializationErrorString, getOneofFieldDescriptor, getRepeatedField, getRepeatedFieldCount, getUnknownFields, hasField, hasOneof
-
Method Details
-
hasScalingMethod
boolean hasScalingMethod()optional .operations_research.glop.GlopParameters.ScalingAlgorithm scaling_method = 57 [default = EQUILIBRATION];
- Returns:
- Whether the scalingMethod field is set.
-
getScalingMethod
GlopParameters.ScalingAlgorithm getScalingMethod()optional .operations_research.glop.GlopParameters.ScalingAlgorithm scaling_method = 57 [default = EQUILIBRATION];
- Returns:
- The scalingMethod.
-
hasFeasibilityRule
boolean hasFeasibilityRule()PricingRule to use during the feasibility phase.
optional .operations_research.glop.GlopParameters.PricingRule feasibility_rule = 1 [default = STEEPEST_EDGE];
- Returns:
- Whether the feasibilityRule field is set.
-
getFeasibilityRule
GlopParameters.PricingRule getFeasibilityRule()PricingRule to use during the feasibility phase.
optional .operations_research.glop.GlopParameters.PricingRule feasibility_rule = 1 [default = STEEPEST_EDGE];
- Returns:
- The feasibilityRule.
-
hasOptimizationRule
boolean hasOptimizationRule()PricingRule to use during the optimization phase.
optional .operations_research.glop.GlopParameters.PricingRule optimization_rule = 2 [default = STEEPEST_EDGE];
- Returns:
- Whether the optimizationRule field is set.
-
getOptimizationRule
GlopParameters.PricingRule getOptimizationRule()PricingRule to use during the optimization phase.
optional .operations_research.glop.GlopParameters.PricingRule optimization_rule = 2 [default = STEEPEST_EDGE];
- Returns:
- The optimizationRule.
-
hasRefactorizationThreshold
boolean hasRefactorizationThreshold()We estimate the factorization accuracy of B during each pivot by using the fact that we can compute the pivot coefficient in two ways: - From direction[leaving_row]. - From update_row[entering_column]. If the two values have a relative difference above this threshold, we trigger a refactorization.
optional double refactorization_threshold = 6 [default = 1e-09];
- Returns:
- Whether the refactorizationThreshold field is set.
-
getRefactorizationThreshold
double getRefactorizationThreshold()We estimate the factorization accuracy of B during each pivot by using the fact that we can compute the pivot coefficient in two ways: - From direction[leaving_row]. - From update_row[entering_column]. If the two values have a relative difference above this threshold, we trigger a refactorization.
optional double refactorization_threshold = 6 [default = 1e-09];
- Returns:
- The refactorizationThreshold.
-
hasRecomputeReducedCostsThreshold
boolean hasRecomputeReducedCostsThreshold()We estimate the accuracy of the iteratively computed reduced costs. If it falls below this threshold, we reinitialize them from scratch. Note that such an operation is pretty fast, so we can use a low threshold. It is important to have a good accuracy here (better than the dual_feasibility_tolerance below) to be sure of the sign of such a cost.
optional double recompute_reduced_costs_threshold = 8 [default = 1e-08];
- Returns:
- Whether the recomputeReducedCostsThreshold field is set.
-
getRecomputeReducedCostsThreshold
double getRecomputeReducedCostsThreshold()We estimate the accuracy of the iteratively computed reduced costs. If it falls below this threshold, we reinitialize them from scratch. Note that such an operation is pretty fast, so we can use a low threshold. It is important to have a good accuracy here (better than the dual_feasibility_tolerance below) to be sure of the sign of such a cost.
optional double recompute_reduced_costs_threshold = 8 [default = 1e-08];
- Returns:
- The recomputeReducedCostsThreshold.
-
hasRecomputeEdgesNormThreshold
boolean hasRecomputeEdgesNormThreshold()Note that the threshold is a relative error on the actual norm (not the squared one) and that edge norms are always greater than 1. Recomputing norms is a really expensive operation and a large threshold is ok since this doesn't impact directly the solution but just the entering variable choice.
optional double recompute_edges_norm_threshold = 9 [default = 100];
- Returns:
- Whether the recomputeEdgesNormThreshold field is set.
-
getRecomputeEdgesNormThreshold
double getRecomputeEdgesNormThreshold()Note that the threshold is a relative error on the actual norm (not the squared one) and that edge norms are always greater than 1. Recomputing norms is a really expensive operation and a large threshold is ok since this doesn't impact directly the solution but just the entering variable choice.
optional double recompute_edges_norm_threshold = 9 [default = 100];
- Returns:
- The recomputeEdgesNormThreshold.
-
hasPrimalFeasibilityTolerance
boolean hasPrimalFeasibilityTolerance()This tolerance indicates by how much we allow the variable values to go out of bounds and still consider the current solution primal-feasible. We also use the same tolerance for the error A.x - b. Note that the two errors are closely related if A is scaled in such a way that the greatest coefficient magnitude on each column is 1.0. This is also simply called feasibility tolerance in other solvers.
optional double primal_feasibility_tolerance = 10 [default = 1e-08];
- Returns:
- Whether the primalFeasibilityTolerance field is set.
-
getPrimalFeasibilityTolerance
double getPrimalFeasibilityTolerance()This tolerance indicates by how much we allow the variable values to go out of bounds and still consider the current solution primal-feasible. We also use the same tolerance for the error A.x - b. Note that the two errors are closely related if A is scaled in such a way that the greatest coefficient magnitude on each column is 1.0. This is also simply called feasibility tolerance in other solvers.
optional double primal_feasibility_tolerance = 10 [default = 1e-08];
- Returns:
- The primalFeasibilityTolerance.
-
hasDualFeasibilityTolerance
boolean hasDualFeasibilityTolerance()Variables whose reduced costs have an absolute value smaller than this tolerance are not considered as entering candidates. That is they do not take part in deciding whether a solution is dual-feasible or not. Note that this value can temporarily increase during the execution of the algorithm if the estimated precision of the reduced costs is higher than this tolerance. Note also that we scale the costs (in the presolve step) so that the cost magnitude range contains one. This is also known as the optimality tolerance in other solvers.
optional double dual_feasibility_tolerance = 11 [default = 1e-08];
- Returns:
- Whether the dualFeasibilityTolerance field is set.
-
getDualFeasibilityTolerance
double getDualFeasibilityTolerance()Variables whose reduced costs have an absolute value smaller than this tolerance are not considered as entering candidates. That is they do not take part in deciding whether a solution is dual-feasible or not. Note that this value can temporarily increase during the execution of the algorithm if the estimated precision of the reduced costs is higher than this tolerance. Note also that we scale the costs (in the presolve step) so that the cost magnitude range contains one. This is also known as the optimality tolerance in other solvers.
optional double dual_feasibility_tolerance = 11 [default = 1e-08];
- Returns:
- The dualFeasibilityTolerance.
-
hasRatioTestZeroThreshold
boolean hasRatioTestZeroThreshold()During the primal simplex (resp. dual simplex), the coefficients of the direction (resp. update row) with a magnitude lower than this threshold are not considered during the ratio test. This tolerance is related to the precision at which a Solve() involving the basis matrix can be performed. TODO(user): Automatically increase it when we detect that the precision of the Solve() is worse than this.
optional double ratio_test_zero_threshold = 12 [default = 1e-09];
- Returns:
- Whether the ratioTestZeroThreshold field is set.
-
getRatioTestZeroThreshold
double getRatioTestZeroThreshold()During the primal simplex (resp. dual simplex), the coefficients of the direction (resp. update row) with a magnitude lower than this threshold are not considered during the ratio test. This tolerance is related to the precision at which a Solve() involving the basis matrix can be performed. TODO(user): Automatically increase it when we detect that the precision of the Solve() is worse than this.
optional double ratio_test_zero_threshold = 12 [default = 1e-09];
- Returns:
- The ratioTestZeroThreshold.
-
hasHarrisToleranceRatio
boolean hasHarrisToleranceRatio()This impacts the ratio test and indicates by how much we allow a basic variable value that we move to go out of bounds. The value should be in [0.0, 1.0) and should be interpreted as a ratio of the primal_feasibility_tolerance. Setting this to 0.0 basically disables the Harris ratio test while setting this too close to 1.0 will make it difficult to keep the variable values inside their bounds modulo the primal_feasibility_tolerance. Note that the same comment applies to the dual simplex ratio test. There, we allow the reduced costs to be of an infeasible sign by as much as this ratio times the dual_feasibility_tolerance.
optional double harris_tolerance_ratio = 13 [default = 0.5];
- Returns:
- Whether the harrisToleranceRatio field is set.
-
getHarrisToleranceRatio
double getHarrisToleranceRatio()This impacts the ratio test and indicates by how much we allow a basic variable value that we move to go out of bounds. The value should be in [0.0, 1.0) and should be interpreted as a ratio of the primal_feasibility_tolerance. Setting this to 0.0 basically disables the Harris ratio test while setting this too close to 1.0 will make it difficult to keep the variable values inside their bounds modulo the primal_feasibility_tolerance. Note that the same comment applies to the dual simplex ratio test. There, we allow the reduced costs to be of an infeasible sign by as much as this ratio times the dual_feasibility_tolerance.
optional double harris_tolerance_ratio = 13 [default = 0.5];
- Returns:
- The harrisToleranceRatio.
-
hasSmallPivotThreshold
boolean hasSmallPivotThreshold()When we choose the leaving variable, we want to avoid small pivot because they are the less precise and may cause numerical instabilities. For a pivot under this threshold times the infinity norm of the direction, we try various countermeasures in order to avoid using it.
optional double small_pivot_threshold = 14 [default = 1e-06];
- Returns:
- Whether the smallPivotThreshold field is set.
-
getSmallPivotThreshold
double getSmallPivotThreshold()When we choose the leaving variable, we want to avoid small pivot because they are the less precise and may cause numerical instabilities. For a pivot under this threshold times the infinity norm of the direction, we try various countermeasures in order to avoid using it.
optional double small_pivot_threshold = 14 [default = 1e-06];
- Returns:
- The smallPivotThreshold.
-
hasMinimumAcceptablePivot
boolean hasMinimumAcceptablePivot()We never follow a basis change with a pivot under this threshold.
optional double minimum_acceptable_pivot = 15 [default = 1e-06];
- Returns:
- Whether the minimumAcceptablePivot field is set.
-
getMinimumAcceptablePivot
double getMinimumAcceptablePivot()We never follow a basis change with a pivot under this threshold.
optional double minimum_acceptable_pivot = 15 [default = 1e-06];
- Returns:
- The minimumAcceptablePivot.
-
hasDropTolerance
boolean hasDropTolerance()In order to increase the sparsity of the manipulated vectors, floating point values with a magnitude smaller than this parameter are set to zero (only in some places). This parameter should be positive or zero.
optional double drop_tolerance = 52 [default = 1e-14];
- Returns:
- Whether the dropTolerance field is set.
-
getDropTolerance
double getDropTolerance()In order to increase the sparsity of the manipulated vectors, floating point values with a magnitude smaller than this parameter are set to zero (only in some places). This parameter should be positive or zero.
optional double drop_tolerance = 52 [default = 1e-14];
- Returns:
- The dropTolerance.
-
hasUseScaling
boolean hasUseScaling()Whether or not we scale the matrix A so that the maximum coefficient on each line and each column is 1.0.
optional bool use_scaling = 16 [default = true];
- Returns:
- Whether the useScaling field is set.
-
getUseScaling
boolean getUseScaling()Whether or not we scale the matrix A so that the maximum coefficient on each line and each column is 1.0.
optional bool use_scaling = 16 [default = true];
- Returns:
- The useScaling.
-
hasCostScaling
boolean hasCostScaling()optional .operations_research.glop.GlopParameters.CostScalingAlgorithm cost_scaling = 60 [default = CONTAIN_ONE_COST_SCALING];
- Returns:
- Whether the costScaling field is set.
-
getCostScaling
GlopParameters.CostScalingAlgorithm getCostScaling()optional .operations_research.glop.GlopParameters.CostScalingAlgorithm cost_scaling = 60 [default = CONTAIN_ONE_COST_SCALING];
- Returns:
- The costScaling.
-
hasInitialBasis
boolean hasInitialBasis()What heuristic is used to try to replace the fixed slack columns in the initial basis of the primal simplex.
optional .operations_research.glop.GlopParameters.InitialBasisHeuristic initial_basis = 17 [default = TRIANGULAR];
- Returns:
- Whether the initialBasis field is set.
-
getInitialBasis
GlopParameters.InitialBasisHeuristic getInitialBasis()What heuristic is used to try to replace the fixed slack columns in the initial basis of the primal simplex.
optional .operations_research.glop.GlopParameters.InitialBasisHeuristic initial_basis = 17 [default = TRIANGULAR];
- Returns:
- The initialBasis.
-
hasUseTransposedMatrix
boolean hasUseTransposedMatrix()Whether or not we keep a transposed version of the matrix A to speed-up the pricing at the cost of extra memory and the initial tranposition computation.
optional bool use_transposed_matrix = 18 [default = true];
- Returns:
- Whether the useTransposedMatrix field is set.
-
getUseTransposedMatrix
boolean getUseTransposedMatrix()Whether or not we keep a transposed version of the matrix A to speed-up the pricing at the cost of extra memory and the initial tranposition computation.
optional bool use_transposed_matrix = 18 [default = true];
- Returns:
- The useTransposedMatrix.
-
hasBasisRefactorizationPeriod
boolean hasBasisRefactorizationPeriod()Number of iterations between two basis refactorizations. Note that various conditions in the algorithm may trigger a refactorization before this period is reached. Set this to 0 if you want to refactorize at each step.
optional int32 basis_refactorization_period = 19 [default = 64];
- Returns:
- Whether the basisRefactorizationPeriod field is set.
-
getBasisRefactorizationPeriod
int getBasisRefactorizationPeriod()Number of iterations between two basis refactorizations. Note that various conditions in the algorithm may trigger a refactorization before this period is reached. Set this to 0 if you want to refactorize at each step.
optional int32 basis_refactorization_period = 19 [default = 64];
- Returns:
- The basisRefactorizationPeriod.
-
hasDynamicallyAdjustRefactorizationPeriod
boolean hasDynamicallyAdjustRefactorizationPeriod()If this is true, then basis_refactorization_period becomes a lower bound on the number of iterations between two refactorization (provided there is no numerical accuracy issues). Depending on the estimated time to refactorize vs the extra time spend in each solves because of the LU update, we try to balance the two times.
optional bool dynamically_adjust_refactorization_period = 63 [default = true];
- Returns:
- Whether the dynamicallyAdjustRefactorizationPeriod field is set.
-
getDynamicallyAdjustRefactorizationPeriod
boolean getDynamicallyAdjustRefactorizationPeriod()If this is true, then basis_refactorization_period becomes a lower bound on the number of iterations between two refactorization (provided there is no numerical accuracy issues). Depending on the estimated time to refactorize vs the extra time spend in each solves because of the LU update, we try to balance the two times.
optional bool dynamically_adjust_refactorization_period = 63 [default = true];
- Returns:
- The dynamicallyAdjustRefactorizationPeriod.
-
hasSolveDualProblem
boolean hasSolveDualProblem()Whether or not we solve the dual of the given problem. With a value of auto, the algorithm decide which approach is probably the fastest depending on the problem dimensions (see dualizer_threshold).
optional .operations_research.glop.GlopParameters.SolverBehavior solve_dual_problem = 20 [default = LET_SOLVER_DECIDE];
- Returns:
- Whether the solveDualProblem field is set.
-
getSolveDualProblem
GlopParameters.SolverBehavior getSolveDualProblem()Whether or not we solve the dual of the given problem. With a value of auto, the algorithm decide which approach is probably the fastest depending on the problem dimensions (see dualizer_threshold).
optional .operations_research.glop.GlopParameters.SolverBehavior solve_dual_problem = 20 [default = LET_SOLVER_DECIDE];
- Returns:
- The solveDualProblem.
-
hasDualizerThreshold
boolean hasDualizerThreshold()When solve_dual_problem is LET_SOLVER_DECIDE, take the dual if the number of constraints of the problem is more than this threshold times the number of variables.
optional double dualizer_threshold = 21 [default = 1.5];
- Returns:
- Whether the dualizerThreshold field is set.
-
getDualizerThreshold
double getDualizerThreshold()When solve_dual_problem is LET_SOLVER_DECIDE, take the dual if the number of constraints of the problem is more than this threshold times the number of variables.
optional double dualizer_threshold = 21 [default = 1.5];
- Returns:
- The dualizerThreshold.
-
hasSolutionFeasibilityTolerance
boolean hasSolutionFeasibilityTolerance()When the problem status is OPTIMAL, we check the optimality using this relative tolerance and change the status to IMPRECISE if an issue is detected. The tolerance is "relative" in the sense that our thresholds are: - tolerance * max(1.0, abs(bound)) for crossing a given bound. - tolerance * max(1.0, abs(cost)) for an infeasible reduced cost. - tolerance for an infeasible dual value.
optional double solution_feasibility_tolerance = 22 [default = 1e-06];
- Returns:
- Whether the solutionFeasibilityTolerance field is set.
-
getSolutionFeasibilityTolerance
double getSolutionFeasibilityTolerance()When the problem status is OPTIMAL, we check the optimality using this relative tolerance and change the status to IMPRECISE if an issue is detected. The tolerance is "relative" in the sense that our thresholds are: - tolerance * max(1.0, abs(bound)) for crossing a given bound. - tolerance * max(1.0, abs(cost)) for an infeasible reduced cost. - tolerance for an infeasible dual value.
optional double solution_feasibility_tolerance = 22 [default = 1e-06];
- Returns:
- The solutionFeasibilityTolerance.
-
hasProvideStrongOptimalGuarantee
boolean hasProvideStrongOptimalGuarantee()If true, then when the solver returns a solution with an OPTIMAL status, we can guarantee that: - The primal variable are in their bounds. - The dual variable are in their bounds. - If we modify each component of the right-hand side a bit and each component of the objective function a bit, then the pair (primal values, dual values) is an EXACT optimal solution of the perturbed problem. - The modifications above are smaller than the associated tolerances as defined in the comment for solution_feasibility_tolerance (*). (*): This is the only place where the guarantee is not tight since we compute the upper bounds with scalar product of the primal/dual solution and the initial problem coefficients with only double precision. Note that whether or not this option is true, we still check the primal/dual infeasibility and objective gap. However if it is false, we don't move the primal/dual values within their bounds and leave them untouched.
optional bool provide_strong_optimal_guarantee = 24 [default = true];
- Returns:
- Whether the provideStrongOptimalGuarantee field is set.
-
getProvideStrongOptimalGuarantee
boolean getProvideStrongOptimalGuarantee()If true, then when the solver returns a solution with an OPTIMAL status, we can guarantee that: - The primal variable are in their bounds. - The dual variable are in their bounds. - If we modify each component of the right-hand side a bit and each component of the objective function a bit, then the pair (primal values, dual values) is an EXACT optimal solution of the perturbed problem. - The modifications above are smaller than the associated tolerances as defined in the comment for solution_feasibility_tolerance (*). (*): This is the only place where the guarantee is not tight since we compute the upper bounds with scalar product of the primal/dual solution and the initial problem coefficients with only double precision. Note that whether or not this option is true, we still check the primal/dual infeasibility and objective gap. However if it is false, we don't move the primal/dual values within their bounds and leave them untouched.
optional bool provide_strong_optimal_guarantee = 24 [default = true];
- Returns:
- The provideStrongOptimalGuarantee.
-
hasChangeStatusToImprecise
boolean hasChangeStatusToImprecise()If true, the internal API will change the return status to imprecise if the solution does not respect the internal tolerances.
optional bool change_status_to_imprecise = 58 [default = true];
- Returns:
- Whether the changeStatusToImprecise field is set.
-
getChangeStatusToImprecise
boolean getChangeStatusToImprecise()If true, the internal API will change the return status to imprecise if the solution does not respect the internal tolerances.
optional bool change_status_to_imprecise = 58 [default = true];
- Returns:
- The changeStatusToImprecise.
-
hasMaxNumberOfReoptimizations
boolean hasMaxNumberOfReoptimizations()When the solution of phase II is imprecise, we re-run the phase II with the opposite algorithm from that imprecise solution (i.e., if primal or dual simplex was used, we use dual or primal simplex, respectively). We repeat such re-optimization until the solution is precise, or we hit this limit.
optional double max_number_of_reoptimizations = 56 [default = 40];
- Returns:
- Whether the maxNumberOfReoptimizations field is set.
-
getMaxNumberOfReoptimizations
double getMaxNumberOfReoptimizations()When the solution of phase II is imprecise, we re-run the phase II with the opposite algorithm from that imprecise solution (i.e., if primal or dual simplex was used, we use dual or primal simplex, respectively). We repeat such re-optimization until the solution is precise, or we hit this limit.
optional double max_number_of_reoptimizations = 56 [default = 40];
- Returns:
- The maxNumberOfReoptimizations.
-
hasLuFactorizationPivotThreshold
boolean hasLuFactorizationPivotThreshold()Threshold for LU-factorization: for stability reasons, the magnitude of the chosen pivot at a given step is guaranteed to be greater than this threshold times the maximum magnitude of all the possible pivot choices in the same column. The value must be in [0,1].
optional double lu_factorization_pivot_threshold = 25 [default = 0.01];
- Returns:
- Whether the luFactorizationPivotThreshold field is set.
-
getLuFactorizationPivotThreshold
double getLuFactorizationPivotThreshold()Threshold for LU-factorization: for stability reasons, the magnitude of the chosen pivot at a given step is guaranteed to be greater than this threshold times the maximum magnitude of all the possible pivot choices in the same column. The value must be in [0,1].
optional double lu_factorization_pivot_threshold = 25 [default = 0.01];
- Returns:
- The luFactorizationPivotThreshold.
-
hasMaxTimeInSeconds
boolean hasMaxTimeInSeconds()Maximum time allowed in seconds to solve a problem.
optional double max_time_in_seconds = 26 [default = inf];
- Returns:
- Whether the maxTimeInSeconds field is set.
-
getMaxTimeInSeconds
double getMaxTimeInSeconds()Maximum time allowed in seconds to solve a problem.
optional double max_time_in_seconds = 26 [default = inf];
- Returns:
- The maxTimeInSeconds.
-
hasMaxDeterministicTime
boolean hasMaxDeterministicTime()Maximum deterministic time allowed to solve a problem. The deterministic time is more or less correlated to the running time, and its unit should be around the second (at least on a Xeon(R) CPU E5-1650 v2 @ 3.50GHz). TODO(user): Improve the correlation.
optional double max_deterministic_time = 45 [default = inf];
- Returns:
- Whether the maxDeterministicTime field is set.
-
getMaxDeterministicTime
double getMaxDeterministicTime()Maximum deterministic time allowed to solve a problem. The deterministic time is more or less correlated to the running time, and its unit should be around the second (at least on a Xeon(R) CPU E5-1650 v2 @ 3.50GHz). TODO(user): Improve the correlation.
optional double max_deterministic_time = 45 [default = inf];
- Returns:
- The maxDeterministicTime.
-
hasMaxNumberOfIterations
boolean hasMaxNumberOfIterations()Maximum number of simplex iterations to solve a problem. A value of -1 means no limit.
optional int64 max_number_of_iterations = 27 [default = -1];
- Returns:
- Whether the maxNumberOfIterations field is set.
-
getMaxNumberOfIterations
long getMaxNumberOfIterations()Maximum number of simplex iterations to solve a problem. A value of -1 means no limit.
optional int64 max_number_of_iterations = 27 [default = -1];
- Returns:
- The maxNumberOfIterations.
-
hasMarkowitzZlatevParameter
boolean hasMarkowitzZlatevParameter()How many columns do we look at in the Markowitz pivoting rule to find a good pivot. See markowitz.h.
optional int32 markowitz_zlatev_parameter = 29 [default = 3];
- Returns:
- Whether the markowitzZlatevParameter field is set.
-
getMarkowitzZlatevParameter
int getMarkowitzZlatevParameter()How many columns do we look at in the Markowitz pivoting rule to find a good pivot. See markowitz.h.
optional int32 markowitz_zlatev_parameter = 29 [default = 3];
- Returns:
- The markowitzZlatevParameter.
-
hasMarkowitzSingularityThreshold
boolean hasMarkowitzSingularityThreshold()If a pivot magnitude is smaller than this during the Markowitz LU factorization, then the matrix is assumed to be singular. Note that this is an absolute threshold and is not relative to the other possible pivots on the same column (see lu_factorization_pivot_threshold).
optional double markowitz_singularity_threshold = 30 [default = 1e-15];
- Returns:
- Whether the markowitzSingularityThreshold field is set.
-
getMarkowitzSingularityThreshold
double getMarkowitzSingularityThreshold()If a pivot magnitude is smaller than this during the Markowitz LU factorization, then the matrix is assumed to be singular. Note that this is an absolute threshold and is not relative to the other possible pivots on the same column (see lu_factorization_pivot_threshold).
optional double markowitz_singularity_threshold = 30 [default = 1e-15];
- Returns:
- The markowitzSingularityThreshold.
-
hasUseDualSimplex
boolean hasUseDualSimplex()Whether or not we use the dual simplex algorithm instead of the primal.
optional bool use_dual_simplex = 31 [default = false];
- Returns:
- Whether the useDualSimplex field is set.
-
getUseDualSimplex
boolean getUseDualSimplex()Whether or not we use the dual simplex algorithm instead of the primal.
optional bool use_dual_simplex = 31 [default = false];
- Returns:
- The useDualSimplex.
-
hasAllowSimplexAlgorithmChange
boolean hasAllowSimplexAlgorithmChange()During incremental solve, let the solver decide if it use the primal or dual simplex algorithm depending on the current solution and on the new problem. Note that even if this is true, the value of use_dual_simplex still indicates the default algorithm that the solver will use.
optional bool allow_simplex_algorithm_change = 32 [default = false];
- Returns:
- Whether the allowSimplexAlgorithmChange field is set.
-
getAllowSimplexAlgorithmChange
boolean getAllowSimplexAlgorithmChange()During incremental solve, let the solver decide if it use the primal or dual simplex algorithm depending on the current solution and on the new problem. Note that even if this is true, the value of use_dual_simplex still indicates the default algorithm that the solver will use.
optional bool allow_simplex_algorithm_change = 32 [default = false];
- Returns:
- The allowSimplexAlgorithmChange.
-
hasDevexWeightsResetPeriod
boolean hasDevexWeightsResetPeriod()Devex weights will be reset to 1.0 after that number of updates.
optional int32 devex_weights_reset_period = 33 [default = 150];
- Returns:
- Whether the devexWeightsResetPeriod field is set.
-
getDevexWeightsResetPeriod
int getDevexWeightsResetPeriod()Devex weights will be reset to 1.0 after that number of updates.
optional int32 devex_weights_reset_period = 33 [default = 150];
- Returns:
- The devexWeightsResetPeriod.
-
hasUsePreprocessing
boolean hasUsePreprocessing()Whether or not we use advanced preprocessing techniques.
optional bool use_preprocessing = 34 [default = true];
- Returns:
- Whether the usePreprocessing field is set.
-
getUsePreprocessing
boolean getUsePreprocessing()Whether or not we use advanced preprocessing techniques.
optional bool use_preprocessing = 34 [default = true];
- Returns:
- The usePreprocessing.
-
hasUseMiddleProductFormUpdate
boolean hasUseMiddleProductFormUpdate()Whether or not to use the middle product form update rather than the standard eta LU update. The middle form product update should be a lot more efficient (close to the Forrest-Tomlin update, a bit slower but easier to implement). See for more details: Qi Huangfu, J. A. Julian Hall, "Novel update techniques for the revised simplex method", 28 january 2013, Technical Report ERGO-13-0001 http://www.maths.ed.ac.uk/hall/HuHa12/ERGO-13-001.pdf
optional bool use_middle_product_form_update = 35 [default = true];
- Returns:
- Whether the useMiddleProductFormUpdate field is set.
-
getUseMiddleProductFormUpdate
boolean getUseMiddleProductFormUpdate()Whether or not to use the middle product form update rather than the standard eta LU update. The middle form product update should be a lot more efficient (close to the Forrest-Tomlin update, a bit slower but easier to implement). See for more details: Qi Huangfu, J. A. Julian Hall, "Novel update techniques for the revised simplex method", 28 january 2013, Technical Report ERGO-13-0001 http://www.maths.ed.ac.uk/hall/HuHa12/ERGO-13-001.pdf
optional bool use_middle_product_form_update = 35 [default = true];
- Returns:
- The useMiddleProductFormUpdate.
-
hasInitializeDevexWithColumnNorms
boolean hasInitializeDevexWithColumnNorms()Whether we initialize devex weights to 1.0 or to the norms of the matrix columns.
optional bool initialize_devex_with_column_norms = 36 [default = true];
- Returns:
- Whether the initializeDevexWithColumnNorms field is set.
-
getInitializeDevexWithColumnNorms
boolean getInitializeDevexWithColumnNorms()Whether we initialize devex weights to 1.0 or to the norms of the matrix columns.
optional bool initialize_devex_with_column_norms = 36 [default = true];
- Returns:
- The initializeDevexWithColumnNorms.
-
hasExploitSingletonColumnInInitialBasis
boolean hasExploitSingletonColumnInInitialBasis()Whether or not we exploit the singleton columns already present in the problem when we create the initial basis.
optional bool exploit_singleton_column_in_initial_basis = 37 [default = true];
- Returns:
- Whether the exploitSingletonColumnInInitialBasis field is set.
-
getExploitSingletonColumnInInitialBasis
boolean getExploitSingletonColumnInInitialBasis()Whether or not we exploit the singleton columns already present in the problem when we create the initial basis.
optional bool exploit_singleton_column_in_initial_basis = 37 [default = true];
- Returns:
- The exploitSingletonColumnInInitialBasis.
-
hasDualSmallPivotThreshold
boolean hasDualSmallPivotThreshold()Like small_pivot_threshold but for the dual simplex. This is needed because the dual algorithm does not interpret this value in the same way. TODO(user): Clean this up and use the same small pivot detection.
optional double dual_small_pivot_threshold = 38 [default = 0.0001];
- Returns:
- Whether the dualSmallPivotThreshold field is set.
-
getDualSmallPivotThreshold
double getDualSmallPivotThreshold()Like small_pivot_threshold but for the dual simplex. This is needed because the dual algorithm does not interpret this value in the same way. TODO(user): Clean this up and use the same small pivot detection.
optional double dual_small_pivot_threshold = 38 [default = 0.0001];
- Returns:
- The dualSmallPivotThreshold.
-
hasPreprocessorZeroTolerance
boolean hasPreprocessorZeroTolerance()A floating point tolerance used by the preprocessors. This is used for things like detecting if two columns/rows are proportional or if an interval is empty. Note that the preprocessors also use solution_feasibility_tolerance() to detect if a problem is infeasible.
optional double preprocessor_zero_tolerance = 39 [default = 1e-09];
- Returns:
- Whether the preprocessorZeroTolerance field is set.
-
getPreprocessorZeroTolerance
double getPreprocessorZeroTolerance()A floating point tolerance used by the preprocessors. This is used for things like detecting if two columns/rows are proportional or if an interval is empty. Note that the preprocessors also use solution_feasibility_tolerance() to detect if a problem is infeasible.
optional double preprocessor_zero_tolerance = 39 [default = 1e-09];
- Returns:
- The preprocessorZeroTolerance.
-
hasObjectiveLowerLimit
boolean hasObjectiveLowerLimit()The solver will stop as soon as it has proven that the objective is smaller than objective_lower_limit or greater than objective_upper_limit. Depending on the simplex algorithm (primal or dual) and the optimization direction, note that only one bound will be used at the time. Important: The solver does not add any tolerances to these values, and as soon as the objective (as computed by the solver, so with some imprecision) crosses one of these bounds (strictly), the search will stop. It is up to the client to add any tolerance if needed.
optional double objective_lower_limit = 40 [default = -inf];
- Returns:
- Whether the objectiveLowerLimit field is set.
-
getObjectiveLowerLimit
double getObjectiveLowerLimit()The solver will stop as soon as it has proven that the objective is smaller than objective_lower_limit or greater than objective_upper_limit. Depending on the simplex algorithm (primal or dual) and the optimization direction, note that only one bound will be used at the time. Important: The solver does not add any tolerances to these values, and as soon as the objective (as computed by the solver, so with some imprecision) crosses one of these bounds (strictly), the search will stop. It is up to the client to add any tolerance if needed.
optional double objective_lower_limit = 40 [default = -inf];
- Returns:
- The objectiveLowerLimit.
-
hasObjectiveUpperLimit
boolean hasObjectiveUpperLimit()optional double objective_upper_limit = 41 [default = inf];
- Returns:
- Whether the objectiveUpperLimit field is set.
-
getObjectiveUpperLimit
double getObjectiveUpperLimit()optional double objective_upper_limit = 41 [default = inf];
- Returns:
- The objectiveUpperLimit.
-
hasDegenerateMinistepFactor
boolean hasDegenerateMinistepFactor()During a degenerate iteration, the more conservative approach is to do a step of length zero (while shifting the bound of the leaving variable). That is, the variable values are unchanged for the primal simplex or the reduced cost are unchanged for the dual simplex. However, instead of doing a step of length zero, it seems to be better on degenerate problems to do a small positive step. This is what is recommended in the EXPAND procedure described in: P. E. Gill, W. Murray, M. A. Saunders, and M. H. Wright. "A practical anti- cycling procedure for linearly constrained optimization". Mathematical Programming, 45:437\u2013474, 1989. Here, during a degenerate iteration we do a small positive step of this factor times the primal (resp. dual) tolerance. In the primal simplex, this may effectively push variable values (very slightly) further out of their bounds (resp. reduced costs for the dual simplex). Setting this to zero reverts to the more conservative approach of a zero step during degenerate iterations.
optional double degenerate_ministep_factor = 42 [default = 0.01];
- Returns:
- Whether the degenerateMinistepFactor field is set.
-
getDegenerateMinistepFactor
double getDegenerateMinistepFactor()During a degenerate iteration, the more conservative approach is to do a step of length zero (while shifting the bound of the leaving variable). That is, the variable values are unchanged for the primal simplex or the reduced cost are unchanged for the dual simplex. However, instead of doing a step of length zero, it seems to be better on degenerate problems to do a small positive step. This is what is recommended in the EXPAND procedure described in: P. E. Gill, W. Murray, M. A. Saunders, and M. H. Wright. "A practical anti- cycling procedure for linearly constrained optimization". Mathematical Programming, 45:437\u2013474, 1989. Here, during a degenerate iteration we do a small positive step of this factor times the primal (resp. dual) tolerance. In the primal simplex, this may effectively push variable values (very slightly) further out of their bounds (resp. reduced costs for the dual simplex). Setting this to zero reverts to the more conservative approach of a zero step during degenerate iterations.
optional double degenerate_ministep_factor = 42 [default = 0.01];
- Returns:
- The degenerateMinistepFactor.
-
hasRandomSeed
boolean hasRandomSeed()At the beginning of each solve, the random number generator used in some part of the solver is reinitialized to this seed. If you change the random seed, the solver may make different choices during the solving process. Note that this may lead to a different solution, for example a different optimal basis. For some problems, the running time may vary a lot depending on small change in the solving algorithm. Running the solver with different seeds enables to have more robust benchmarks when evaluating new features. Also note that the solver is fully deterministic: two runs of the same binary, on the same machine, on the exact same data and with the same parameters will go through the exact same iterations. If they hit a time limit, they might of course yield different results because one will have advanced farther than the other.
optional int32 random_seed = 43 [default = 1];
- Returns:
- Whether the randomSeed field is set.
-
getRandomSeed
int getRandomSeed()At the beginning of each solve, the random number generator used in some part of the solver is reinitialized to this seed. If you change the random seed, the solver may make different choices during the solving process. Note that this may lead to a different solution, for example a different optimal basis. For some problems, the running time may vary a lot depending on small change in the solving algorithm. Running the solver with different seeds enables to have more robust benchmarks when evaluating new features. Also note that the solver is fully deterministic: two runs of the same binary, on the same machine, on the exact same data and with the same parameters will go through the exact same iterations. If they hit a time limit, they might of course yield different results because one will have advanced farther than the other.
optional int32 random_seed = 43 [default = 1];
- Returns:
- The randomSeed.
-
hasUseAbslRandom
boolean hasUseAbslRandom()Whether to use absl::BitGen instead of MTRandom.
optional bool use_absl_random = 72 [default = false];
- Returns:
- Whether the useAbslRandom field is set.
-
getUseAbslRandom
boolean getUseAbslRandom()Whether to use absl::BitGen instead of MTRandom.
optional bool use_absl_random = 72 [default = false];
- Returns:
- The useAbslRandom.
-
hasNumOmpThreads
boolean hasNumOmpThreads()Number of threads in the OMP parallel sections. If left to 1, the code will not create any OMP threads and will remain single-threaded.
optional int32 num_omp_threads = 44 [default = 1];
- Returns:
- Whether the numOmpThreads field is set.
-
getNumOmpThreads
int getNumOmpThreads()Number of threads in the OMP parallel sections. If left to 1, the code will not create any OMP threads and will remain single-threaded.
optional int32 num_omp_threads = 44 [default = 1];
- Returns:
- The numOmpThreads.
-
hasPerturbCostsInDualSimplex
boolean hasPerturbCostsInDualSimplex()When this is true, then the costs are randomly perturbed before the dual simplex is even started. This has been shown to improve the dual simplex performance. For a good reference, see Huangfu Q (2013) "High performance simplex solver", Ph.D, dissertation, University of Edinburgh.
optional bool perturb_costs_in_dual_simplex = 53 [default = false];
- Returns:
- Whether the perturbCostsInDualSimplex field is set.
-
getPerturbCostsInDualSimplex
boolean getPerturbCostsInDualSimplex()When this is true, then the costs are randomly perturbed before the dual simplex is even started. This has been shown to improve the dual simplex performance. For a good reference, see Huangfu Q (2013) "High performance simplex solver", Ph.D, dissertation, University of Edinburgh.
optional bool perturb_costs_in_dual_simplex = 53 [default = false];
- Returns:
- The perturbCostsInDualSimplex.
-
hasUseDedicatedDualFeasibilityAlgorithm
boolean hasUseDedicatedDualFeasibilityAlgorithm()We have two possible dual phase I algorithms. Both work on an LP that minimize the sum of dual infeasiblities. One use dedicated code (when this param is true), the other one use exactly the same code as the dual phase II but on an auxiliary problem where the variable bounds of the original problem are changed. TODO(user): For now we have both, but ideally the non-dedicated version will win since it is a lot less code to maintain.
optional bool use_dedicated_dual_feasibility_algorithm = 62 [default = true];
- Returns:
- Whether the useDedicatedDualFeasibilityAlgorithm field is set.
-
getUseDedicatedDualFeasibilityAlgorithm
boolean getUseDedicatedDualFeasibilityAlgorithm()We have two possible dual phase I algorithms. Both work on an LP that minimize the sum of dual infeasiblities. One use dedicated code (when this param is true), the other one use exactly the same code as the dual phase II but on an auxiliary problem where the variable bounds of the original problem are changed. TODO(user): For now we have both, but ideally the non-dedicated version will win since it is a lot less code to maintain.
optional bool use_dedicated_dual_feasibility_algorithm = 62 [default = true];
- Returns:
- The useDedicatedDualFeasibilityAlgorithm.
-
hasRelativeCostPerturbation
boolean hasRelativeCostPerturbation()The magnitude of the cost perturbation is given by RandomIn(1.0, 2.0) * ( relative_cost_perturbation * cost + relative_max_cost_perturbation * max_cost);
optional double relative_cost_perturbation = 54 [default = 1e-05];
- Returns:
- Whether the relativeCostPerturbation field is set.
-
getRelativeCostPerturbation
double getRelativeCostPerturbation()The magnitude of the cost perturbation is given by RandomIn(1.0, 2.0) * ( relative_cost_perturbation * cost + relative_max_cost_perturbation * max_cost);
optional double relative_cost_perturbation = 54 [default = 1e-05];
- Returns:
- The relativeCostPerturbation.
-
hasRelativeMaxCostPerturbation
boolean hasRelativeMaxCostPerturbation()optional double relative_max_cost_perturbation = 55 [default = 1e-07];
- Returns:
- Whether the relativeMaxCostPerturbation field is set.
-
getRelativeMaxCostPerturbation
double getRelativeMaxCostPerturbation()optional double relative_max_cost_perturbation = 55 [default = 1e-07];
- Returns:
- The relativeMaxCostPerturbation.
-
hasInitialConditionNumberThreshold
boolean hasInitialConditionNumberThreshold()If our upper bound on the condition number of the initial basis (from our heurisitic or a warm start) is above this threshold, we revert to an all slack basis.
optional double initial_condition_number_threshold = 59 [default = 1e+50];
- Returns:
- Whether the initialConditionNumberThreshold field is set.
-
getInitialConditionNumberThreshold
double getInitialConditionNumberThreshold()If our upper bound on the condition number of the initial basis (from our heurisitic or a warm start) is above this threshold, we revert to an all slack basis.
optional double initial_condition_number_threshold = 59 [default = 1e+50];
- Returns:
- The initialConditionNumberThreshold.
-
hasLogSearchProgress
boolean hasLogSearchProgress()If true, logs the progress of a solve to LOG(INFO). Note that the same messages can also be turned on by displaying logs at level 1 for the relevant files.
optional bool log_search_progress = 61 [default = false];
- Returns:
- Whether the logSearchProgress field is set.
-
getLogSearchProgress
boolean getLogSearchProgress()If true, logs the progress of a solve to LOG(INFO). Note that the same messages can also be turned on by displaying logs at level 1 for the relevant files.
optional bool log_search_progress = 61 [default = false];
- Returns:
- The logSearchProgress.
-
hasLogToStdout
boolean hasLogToStdout()If true, logs will be displayed to stdout instead of using Google log info.
optional bool log_to_stdout = 66 [default = true];
- Returns:
- Whether the logToStdout field is set.
-
getLogToStdout
boolean getLogToStdout()If true, logs will be displayed to stdout instead of using Google log info.
optional bool log_to_stdout = 66 [default = true];
- Returns:
- The logToStdout.
-
hasCrossoverBoundSnappingDistance
boolean hasCrossoverBoundSnappingDistance()If the starting basis contains FREE variable with bounds, we will move any such variable to their closer bounds if the distance is smaller than this parameter. The starting statuses can contains FREE variables with bounds, if a user set it like this externally. Also, any variable with an initial BASIC status that was not kept in the initial basis is marked as FREE before this step is applied. Note that by default a FREE variable is assumed to be zero unless a starting value was specified via SetStartingVariableValuesForNextSolve(). Note that, at the end of the solve, some of these FREE variable with bounds and an interior point value might still be left in the final solution. Enable push_to_vertex to clean these up.
optional double crossover_bound_snapping_distance = 64 [default = inf];
- Returns:
- Whether the crossoverBoundSnappingDistance field is set.
-
getCrossoverBoundSnappingDistance
double getCrossoverBoundSnappingDistance()If the starting basis contains FREE variable with bounds, we will move any such variable to their closer bounds if the distance is smaller than this parameter. The starting statuses can contains FREE variables with bounds, if a user set it like this externally. Also, any variable with an initial BASIC status that was not kept in the initial basis is marked as FREE before this step is applied. Note that by default a FREE variable is assumed to be zero unless a starting value was specified via SetStartingVariableValuesForNextSolve(). Note that, at the end of the solve, some of these FREE variable with bounds and an interior point value might still be left in the final solution. Enable push_to_vertex to clean these up.
optional double crossover_bound_snapping_distance = 64 [default = inf];
- Returns:
- The crossoverBoundSnappingDistance.
-
hasPushToVertex
boolean hasPushToVertex()If the optimization phases finishes with super-basic variables (i.e., variables that either 1) have bounds but are FREE in the basis, or 2) have no bounds and are FREE in the basis at a nonzero value), then run a "push" phase to push these variables to bounds, obtaining a vertex solution. Note this situation can happen only if a starting value was specified via SetStartingVariableValuesForNextSolve().
optional bool push_to_vertex = 65 [default = true];
- Returns:
- Whether the pushToVertex field is set.
-
getPushToVertex
boolean getPushToVertex()If the optimization phases finishes with super-basic variables (i.e., variables that either 1) have bounds but are FREE in the basis, or 2) have no bounds and are FREE in the basis at a nonzero value), then run a "push" phase to push these variables to bounds, obtaining a vertex solution. Note this situation can happen only if a starting value was specified via SetStartingVariableValuesForNextSolve().
optional bool push_to_vertex = 65 [default = true];
- Returns:
- The pushToVertex.
-
hasUseImpliedFreePreprocessor
boolean hasUseImpliedFreePreprocessor()If presolve runs, include the pass that detects implied free variables.
optional bool use_implied_free_preprocessor = 67 [default = true];
- Returns:
- Whether the useImpliedFreePreprocessor field is set.
-
getUseImpliedFreePreprocessor
boolean getUseImpliedFreePreprocessor()If presolve runs, include the pass that detects implied free variables.
optional bool use_implied_free_preprocessor = 67 [default = true];
- Returns:
- The useImpliedFreePreprocessor.
-
hasMaxValidMagnitude
boolean hasMaxValidMagnitude()Any finite values in the input LP must be below this threshold, otherwise the model will be reported invalid. This is needed to avoid floating point overflow when evaluating bounds * coeff for instance. In practice, users shouldn't use super large values in an LP. With the default threshold, even evaluating large constraint with variables at their bound shouldn't cause any overflow.
optional double max_valid_magnitude = 70 [default = 1e+30];
- Returns:
- Whether the maxValidMagnitude field is set.
-
getMaxValidMagnitude
double getMaxValidMagnitude()Any finite values in the input LP must be below this threshold, otherwise the model will be reported invalid. This is needed to avoid floating point overflow when evaluating bounds * coeff for instance. In practice, users shouldn't use super large values in an LP. With the default threshold, even evaluating large constraint with variables at their bound shouldn't cause any overflow.
optional double max_valid_magnitude = 70 [default = 1e+30];
- Returns:
- The maxValidMagnitude.
-
hasDropMagnitude
boolean hasDropMagnitude()Value in the input LP lower than this will be ignored. This is similar to drop_tolerance but more aggressive as this is used before scaling. This is mainly here to avoid underflow and have simpler invariant in the code, like a * b == 0 iff a or b is zero and things like this.
optional double drop_magnitude = 71 [default = 1e-30];
- Returns:
- Whether the dropMagnitude field is set.
-
getDropMagnitude
double getDropMagnitude()Value in the input LP lower than this will be ignored. This is similar to drop_tolerance but more aggressive as this is used before scaling. This is mainly here to avoid underflow and have simpler invariant in the code, like a * b == 0 iff a or b is zero and things like this.
optional double drop_magnitude = 71 [default = 1e-30];
- Returns:
- The dropMagnitude.
-
hasDualPricePrioritizeNorm
boolean hasDualPricePrioritizeNorm()On some problem like stp3d or pds-100 this makes a huge difference in speed and number of iterations of the dual simplex.
optional bool dual_price_prioritize_norm = 69 [default = false];
- Returns:
- Whether the dualPricePrioritizeNorm field is set.
-
getDualPricePrioritizeNorm
boolean getDualPricePrioritizeNorm()On some problem like stp3d or pds-100 this makes a huge difference in speed and number of iterations of the dual simplex.
optional bool dual_price_prioritize_norm = 69 [default = false];
- Returns:
- The dualPricePrioritizeNorm.
-