def step(self):
pop_size = len(self.population)
for i in range(pop_size):
specimen_pos = self.population[i]
specimen_best_pos = self.bestPositions[i]
new_speed = []
for att_i in range(self.d):
local_best_tend = self.c1 * rand.random() * (specimen_best_pos[att_i] - (specimen_pos[att_i]))
global_best_tend = self.c2 * rand.random() * (self.gBest[att_i] - (specimen_pos[att_i]))
new_speed.append(self.w * self.actualSpeed[i][att_i] + local_best_tend + global_best_tend)
# speed validation and correction
self.validate_speed_vector(new_speed)
self.actualSpeed[i] = new_speed
# new position update
new_position = []
for att_i in range(self.d):
new_position.append(specimen_pos[att_i] + new_speed[att_i])
# position validation
PopulationUtils.validate_constrains(new_position, self.specimen_template)
# count fitness and update new position
new_position.append(self.fitness_function(new_position))
self.population[i] = new_position
# control with best local and global
if self.gBest[self.d] > new_position[self.d]:
self.gBest = new_position
if self.bestPositions[i][self.d] > new_position[self.d]:
self.bestPositions[i] = new_position
self.iteration += 1
def step(self):
new_population = []
pop_size = len(self.population)
for i in range(pop_size):
actual_specimen = self.population[i]
# choose 3 random specimen from population
random_specimens = []
for j in range(3):
next_spec_index = rand.randint(0, pop_size-1)
next_spec = self.population[next_spec_index]
# if we pick already used, we just increase index until we find unused
while next_spec in random_specimens:
next_spec_index += 1
if next_spec_index == pop_size:
next_spec_index = 0
next_spec = self.population[next_spec_index]
# now we have unused - use numpy arrays for easy manipulation
random_specimens.append(np.array(next_spec[:self.d]))
# create Differential vector
differential_vector = np.array(random_specimens[0] - random_specimens[1])
weighted_differential_vector = self.F * differential_vector
noise_vector = random_specimens[2] + weighted_differential_vector
# trial vector - MUTATION part
trial_vector = []
for att_index in range(len(noise_vector)):
probability = rand.random()
if probability <= self.CR:
trial_vector.append(noise_vector[att_index])
else:
trial_vector.append(actual_specimen[att_index])
# constrains
PopulationUtils.validate_constrains(trial_vector, self.specimen_template)
# count fitness of trial vector
trial_vector.append(self.fitness_function(trial_vector))
# add to new population one with better fitness
if trial_vector[self.d] < actual_specimen[self.d]:
new_population.append(trial_vector)
else:
new_population.append(actual_specimen)
# discard old population in favour of new one
self.population = new_population
# or generation to be correct
self.iteration += 1
def neighborhood(x, d, fitness_fn=None, n=10, specimen_template=None):
"""
Generates randomly neighbourhood of specimen "x" in given diameter "d"
:param x: specimen
:param d: area diameter
:param fitness_fn: fitness function
:param specimen_template: template
:param n: number of neighbours to generate default=10
:return: list of neighbours
"""
neighbours = []
for i in range(n):
new = []
for dimension in x:
# (dimension-d) to always be in parenthesis
new.append((rand.random() * (dimension+d - (dimension-d))) + (dimension-d))
if specimen_template is not None:
PopulationUtils.validate_constrains(new, specimen_template)
if fitness_fn is not None:
new.append(fitness_fn(new))
neighbours.append(new)
return neighbours
