Solver Algorithms Options

DEPS Evolutionary Algorithm

DEPS consists of two independent algorithms: Differential Evolution and Particle Swarm Optimization. Both are especially suited for numerical problems, such as nonlinear optimization, and are complementary to each other in that they even out each other’s shortcomings.

Setting	Description
Agent Switch Rate	Specifies the probability for an individual to choose the Differential Evolution strategy.
Assume variables as non negative	Mark to force variables to be positive only.
DE: Crossover Probability	Defines the probability of the individual being combined with the globally best point. If crossover is not used, the point is assembled from the own memory of the individual.
DE: Scaling Factor	During crossover, the scaling factor decides about the “speed” of movement.
Learning Cycles	Defines the number of iterations, the algorithm should take. In each iteration, all individuals make a guess on the best solution and share their knowledge.
PS: Cognitive Constant	Sets the importance of the own memory (in particular the best reached point so far).
PS: Constriction Coefficient	Defines the speed at which the particles/individuals move towards each other.
PS: Mutation Probability	Defines the probability, that instead of moving a component of the particle towards the best point, it randomly chooses a new value from the valid range for that variable.
PS: Social Constant	Sets the importance of the global best point between all particles/individuals.
Show Enhanced Solver Status	If enabled, an additional dialog is shown during the solving process which gives information about the current progress, the level of stagnation, the currently best known solution as well as the possibility, to stop or resume the solver.
Size of Swarm	Defines the number of individuals to participate in the learning process. Each individual finds its own solutions and contributes to the overall knowledge.
Stagnation Limit	If this number of individuals found solutions within a close range, the iteration is stopped and the best of these values is chosen as optimal.
Stagnation Tolerance	Defines in what range solutions are considered “similar”.
Use ACR Comparator	If disabled (default), the BCH Comparator is used. It compares two individuals by first looking at their constraint violations and only if those are equal, it measures their current solution. If enabled, the ACR Comparator is used. It compares two individuals dependent on the current iteration and measures their goodness with knowledge about the libraries worst known solutions (in regard to their constraint violations).
Use Random Starting Point	If enabled, the library is simply filled up with randomly chosen points. If disabled, the currently present values (as given by the user) are inserted in the library as reference point.
Variable Bounds Guessing	If enabled (default), the algorithm tries to find variable bounds by looking at the starting values.
Variable Bounds Threshold	When guessing variable bounds, this threshold specifies, how the initial values are shifted to build the bounds. For an example how these values are calculated, please refer to the Manual in the Wiki.

SCO Evolutionary Algorithm

Social Cognitive Optimization takes into account the human behavior of learning and sharing information. Each individual has access to a common library with knowledge shared between all individuals.

Setting	Description
Assume variables as non negative	Mark to force variables to be positive only.
Learning Cycles	Defines the number of iterations, the algorithm should take. In each iteration, all individuals make a guess on the best solution and share their knowledge.
Show Enhanced Solver Status	If enabled, an additional dialog is shown during the solving process which gives information about the current progress, the level of stagnation, the currently best known solution as well as the possibility, to stop or resume the solver.
Size of Library	Defines the amount of information to store in the public library. Each individual stores knowledge there and asks for information.
Size of Swarm	Defines the number of individuals to participate in the learning process. Each individual finds its own solutions and contributes to the overall knowledge.
Stagnation Limit	If this number of individuals found solutions within a close range, the iteration is stopped and the best of these values is chosen as optimal.
Stagnation Tolerance	Defines in what range solutions are considered “similar”.
Use ACR Comparator	If disabled (default), the BCH Comparator is used. It compares two individuals by first looking at their constraint violations and only if those are equal, it measures their current solution. If enabled, the ACR Comparator is used. It compares two individuals dependent on the current iteration and measures their goodness with knowledge about the libraries worst known solutions (in regard to their constraint violations).
Variable Bounds Guessing	If enabled (default), the algorithm tries to find variable bounds by looking at the starting values.
Variable Bounds Threshold	When guessing variable bounds, this threshold specifies, how the initial values are shifted to build the bounds. For an example how these values are calculated, please refer to the Manual in the Wiki.

LibreOffice Linear Solver and CoinMP Linear solver

Setting	Description
Assume variables as integers	Mark to force variables to be integers only.
Assume variables as non negative	Mark to force variables to be positive only.
Epsilon level	Epsilon level. Valid values are in range 0 (very tight) to 3 (very loose). Epsilon is the tolerance for rounding values to zero.
Limit branch-and-bound depth	Specifies the maximum branch-and-bound depth. A positive value means that the depth is absolute. A negative value means a relative branch-and-bound depth limit.
Solver time limit	Sets the maximum time for the algorithm to converge to a solution.

LibreOffice Swarm Non-Linear Solver (Experimental)

Setting	Description
Assume variables as integers	Mark to force variables to be integers only.
Assume variables as non negative	Mark to force variables to be positive only.
Solver time limit	Sets the maximum time for the algorithm to converge to a solution.
Swarm algorithm	Set the swarm algorithm. 0 for differential evolution and 1 for particle swarm optimization. Default is 0.

Solver Options

Solver

Wiki page on non-linear solvers and their algorithms

lp_solve reference guide in sourceforge.net