culebra.trainer.aco
module¶
Implementation of some ant colony optimization trainers.
This module is composed by:
The
abc
sub-module, where some abstract base classes are defined to support the ACO trainers developed in this moduleSome popular single-objective ACO algorithms:
The
AntSystem
class, which implements the Ant System algorithmThe
ElitistAntSystem
class, which implements the Elitist Ant System algorithmThe
MMAS
class, which implements the \({\small \mathcal{MAX}{-}\mathcal{MIN}}\) Ant System algorithmThe
AgeBasedPACO
class, which implements a PACO approach with an age-based population update strategyThe
QualityBasedPACO
class, which implements a PACO approach with a quality-based population update strategySome multi-objective ACO algorithms:
ACO-based wrappers for Feature Selection problems:
The
PACO_FS
class. A new population-based wrapper ACO approach for Feature Selection problemsThe
ElitistACO_FS
class. A new elitist ACO-based wrapper for Feature Selection problemsThe
ACO_FSConvergenceDetector
class, designed to enable stoppingACO_FS
trainers upon convergence
Attributes¶
- DEFAULT_PHEROMONE_INFLUENCE = 1.0¶
Default pheromone influence (\({\alpha}\)).
- DEFAULT_HEURISTIC_INFLUENCE = 2.0¶
Default heuristic influence (\({\beta}\)).
- DEFAULT_CONVERGENCE_CHECK_FREQ = 100¶
Default frequency to check if an elitist ACO has converged.
- DEFAULT_PHEROMONE_EVAPORATION_RATE = 0.1¶
Default pheromone evaporation rate (\({\rho}\)).
- DEFAULT_ELITE_WEIGHT = 0.3¶
Default weight for the elite ant (best-so-far ant) respect to the iteration-best ant.
- DEFAULT_MMAS_ITER_BEST_USE_LIMIT = 250¶
Default limit for the number of iterations for the \({\small \mathcal{MAX}{-}\mathcal{MIN}}\) AS to give up using the iteration-best ant to deposit pheromone. Iterations above this limit will use only the global-best ant.
- DEFAULT_ACO_FS_INITIAL_PHEROMONE = 1¶
Default initial pheromone.
- DEFAULT_ACO_FS_DISCARD_PROB = 0.5¶
Default probability of discarding a node (feature).