def __init__(self, problem, parallel=False, *args, **kwargs):
"""
Create object instance and perform a single optimization run using PSO.
:param problem: The problem to be optimized. Instance of Problem-class.
:param parallel:
Evaluate the fitness for the particles in parallel.
See the README.md file for a discussion on this.
:return:
Object instance. Get the optimization results from the object's variables.
- best is the best-found solution.
- best_fitness is the associated fitness of the best-found solution.
- fitness_trace is an instance of the FitnessTrace-class.
"""
# Copy arguments to instance variables.
self.problem = problem
self.parallel = parallel
# Initialize all particles with random positions in the search-space.
# The first index is for the particle number.
# The second index is for the search-space dimension.
# Note that self.num_particles must be set prior to this by the sub-class.
self.particle = tools.rand_population(lower=problem.lower_init,
upper=problem.upper_init,
num_agents=self.num_particles,
dim=problem.dim)
# Initialize best-known positions for the particles to their starting positions.
# A copy is made because the particle positions will change during optimization
# regardless of improvement to the particle's fitness.
self.particle_best = np.copy(self.particle)
# Initialize fitness of best-known particle positions to infinity.
self.particle_best_fitness = np.repeat(np.inf, self.num_particles)
# Boundaries for the velocity. These are set to the range of the search-space.
bound_range = np.abs(problem.upper_bound - problem.lower_bound)
self.velocity_lower_bound = -bound_range
self.velocity_upper_bound = bound_range
# Initialize all velocities with random values in the allowed range.
self.velocity = tools.rand_population(lower=self.velocity_lower_bound,
upper=self.velocity_upper_bound,
num_agents=self.num_particles,
dim=problem.dim)
# Initialize parent-class which also starts the optimization run.
SingleRun.__init__(self, *args, **kwargs)
def __init__(self, problem, parallel=False, *args, **kwargs):
"""
Create object instance and perform a single optimization run using PSO.
:param problem: The problem to be optimized. Instance of Problem-class.
:param parallel:
Evaluate the fitness for the particles in parallel.
See the README.md file for a discussion on this.
:return:
Object instance. Get the optimization results from the object's variables.
- best is the best-found solution.
- best_fitness is the associated fitness of the best-found solution.
- fitness_trace is an instance of the FitnessTrace-class.
"""
# Copy arguments to instance variables.
self.problem = problem
self.parallel = parallel
# Initialize all particles with random positions in the search-space.
# The first index is for the particle number.
# The second index is for the search-space dimension.
# Note that self.num_particles must be set prior to this by the sub-class.
self.particle = tools.rand_population(lower=problem.lower_init,
upper=problem.upper_init,
num_agents=self.num_particles,
dim=problem.dim)
# Initialize best-known positions for the particles to their starting positions.
# A copy is made because the particle positions will change during optimization
# regardless of improvement to the particle's fitness.
self.particle_best = np.copy(self.particle)
# Initialize fitness of best-known particle positions to infinity.
self.particle_best_fitness = np.repeat(np.inf, self.num_particles)
# Boundaries for the velocity. These are set to the range of the search-space.
bound_range = np.abs(problem.upper_bound - problem.lower_bound)
self.velocity_lower_bound = -bound_range
self.velocity_upper_bound = bound_range
# Initialize all velocities with random values in the allowed range.
self.velocity = tools.rand_population(lower=self.velocity_lower_bound,
upper=self.velocity_upper_bound,
num_agents=self.num_particles,
dim=problem.dim)
# Initialize parent-class which also starts the optimization run.
SingleRun.__init__(self, *args, **kwargs)
def __init__(self, problem, parameters=parameters_default, parallel=False, *args, **kwargs):
"""
Create object instance and perform a single optimization run using DE.
:param problem: The problem to be optimized. Instance of Problem-class.
:param parameters:
Control parameters for the DE.
These may have a significant impact on the optimization performance.
First try and use the default parameters and if they don't give satisfactory
results, then experiment with other parameters or meta-optimization.
:param parallel:
Evaluate the fitness for the agents in parallel.
See the README.md file for a discussion on this.
:return:
Object instance. Get the optimization results from the object's variables.
- best is the best-found solution.
- best_fitness is the associated fitness of the best-found solution.
- fitness_trace is an instance of the FitnessTrace-class.
"""
# Copy arguments to instance variables.
self.problem = problem
self.parallel = parallel
# Unpack control parameters.
self.num_agents, self.crossover_probability, self.differential_weight = parameters
# The number of agents must be an integer.
self.num_agents = int(self.num_agents)
# Initialize population with random positions in the search-space.
# The first index is for the agent number.
# The second index is for the search-space dimension.
# These are actually the best-known positions for each agent
# and will only get replaced when improvements are found.
self.population = tools.rand_population(lower=problem.lower_init,
upper=problem.upper_init,
num_agents=self.num_agents,
dim=problem.dim)
# Population used for generating new agents.
self.new_population = np.copy(self.population)
# Initialize fitness for all agents to infinity.
self.fitness = np.repeat(np.inf, self.num_agents)
# Initialize parent-class which also starts the optimization run.
SingleRun.__init__(self, *args, **kwargs)