def initialize(self):
# set the objective function
self.default_vals["val_size"] = 30
self.default_vals["weight_acc"] = 0.9
if self.default_mode:
self.val_size = self.default_vals["val_size"]
self.weight_acc = self.default_vals["weight_acc"]
else:
self.val_size = float(input(f'Percentage of data for valdiation [0-100] (default={self.default_vals["val_size"]}): ') or self.default_vals["val_size"])/100
self.weight_acc = float(input(f'Weight for the classification accuracy [0-1] (default={self.default_vals["weight_acc"]}): ') or self.default_vals["weight_acc"])
self.obj_function = call_counter(compute_fitness(self.weight_acc))
# start timer
self.start_time = time.time()
np.random.seed(self.seed)
# data preparation
self.training_data = Data()
self.train_data, self.train_label = np.array(self.train_data), np.array(self.train_label)
self.train_label = self.int_encoding(self.train_label)
self.training_data.train_X, self.training_data.val_X, self.training_data.train_Y, self.training_data.val_Y = train_test_split(self.train_data, self.train_label, stratify=self.train_label, test_size=self.val_size)
# create initial population
self.num_features = self.train_data.shape[1]
self.population = initialize(num_agents=self.num_agents, num_features=self.num_features, seed=self.seed)
self.fitness = self.obj_function(self.population, self.training_data)
self.population, self.fitness = sort_agents(agents=self.population, fitness=self.fitness)
self.accuracy = compute_accuracy(agents=self.population, data=self.training_data)
self.Leader_agent, self.Leader_fitness = self.population[0], self.fitness[0]
def initialize(self):
#call the base class function
super().initialize()
# initialize position and velocities of male and female mayflies' and Leader (the agent with the max fitness)
self.male_pos = initialize(self.num_agents, self.num_features)
self.female_pos = initialize(self.num_agents, self.num_features)
self.male_vel = np.random.uniform(low = -1, high = 1, size = (self.num_agents, self.num_features))
self.female_vel = np.random.uniform(low = -1, high = 1, size = (self.num_agents, self.num_features))
self.male_fitness = np.zeros(self.num_agents)
self.male_accuracy = np.zeros(self.num_agents)
self.female_fitness = np.zeros(self.num_agents)
self.male_personal_best = np.zeros((self.num_agents, self.num_features))
self.male_personal_best_fit = np.zeros(self.num_agents)
self.male_offspring = np.zeros((self.num_agents, self.num_features))
self.female_offspring = np.zeros((self.num_agents, self.num_features))
self.vmax_male = np.zeros(self.num_features)
self.vmax_female = np.zeros(self.num_features)
# rank initial population
self.male_fitness = self.obj_function(self.male_pos,self.training_data)
self.female_fitness = self.obj_function(self.female_pos,self.training_data)
self.male_pos, self.male_fitness = sort_agents(self.male_pos,self.male_fitness)
self.female_pos, self.female_fitness = sort_agents(self.female_pos,self.female_fitness)
# create initial population
self.population = self.male_pos
self.fitness = self.male_fitness
self.population, self.fitness = sort_agents(agents=self.population, fitness=self.fitness)
self.accuracy = compute_accuracy(agents=self.population, data=self.training_data)
self.Leader_agent, self.Leader_fitness = self.population[0], self.fitness[0]
def post_processing(self):
# post processing steps
self.fitness = self.obj_function(self.population, self.training_data)
self.population, self.fitness = sort_agents(agents=self.population, fitness=self.fitness)
self.accuracy = compute_accuracy(agents=self.population, data=self.training_data)
if(self.fitness[0] > self.Leader_fitness):
self.Leader_fitness = self.fitness[0]
self.Leader_agent = self.population[0, :]
self.Leader_accuracy = self.accuracy[0]
def run(self):
# the main algorithm run
if not self.default_mode:
self.print(
'\n************ Please enter the values of the following paramters or press newline for using default values ************\n'
)
self.user_input() # take the user inputs
self.initialize() # initialize the algorithm
self.save_details() # save the initial details
if not self.default_mode:
self.print(
'\n***************************************************** Thank You ******************************************************\n'
)
while (not self.check_end()): # while the end criterion is not met
self.next() # do one step of the algorithm
self.post_processing() # do the post processing steps
self.display() # display the details of 1 iteration
self.save_details() # save the details
self.end_time = time.time()
self.exec_time = self.end_time - self.start_time
if self.test_data: # if there is a test data, test the final solution on that
self.test_label = self.int_encoding(self.test_label)
temp_data = Data()
temp_data.train_X = self.train_data
temp_data.train_Y = self.train_label
temp_data.val_X = self.test_data
temp_data.val_Y = self.test_label
self.Leader_fitness = compute_fitness(self.Leader_agent, temp_data)
self.Leader_accuracy = compute_accuracy(self.Leader_agent,
temp_data)
self.save_solution()
if (self.save_conv_graph):
fig = self.plot()
fig.savefig('convergence_curve_' + self.algo_name + '.jpg')
self.print('\n------------- Leader Agent ---------------')
self.print('Fitness: {}'.format(self.Leader_fitness))
self.print('Number of Features: {}'.format(
int(np.sum(self.Leader_agent))))
self.print('----------------------------------------\n')
return self