evolution.py

import random
import numpy as np
import trace_processor
import glob
import infinite_game
import Levenshtein as lv 
import predictor
import gc

class Evo:
	def __init__(self, population_size, truncation_percentage, mutation_rate, max_generations, cluster_example, map_name):
		# Tuning parameters
		self.population_size = population_size
		self.truncation_selection = int(self.population_size * truncation_percentage)
		self.mutation_rate = mutation_rate

		self.current_generation = 0
		#list of lists of parameters
		self.population = []

		self.map_name = map_name

		self.cluster_example = cluster_example

		self.average_fitness_evolution = []
		self.best_agent_evolution = []
		self.best_fitness_evolution = []




	def get_initial_population(self):
		

		for parameter_num in range(population_size):

			individual = random.choices(range(0, 101), k = 10)

			self.population.append(individual)

		print("Initial Population: ", self.population)
		return




	def get_fitness(self, agent_parameters):

		print("Agent Playing: ", agent_parameters)

		action_list, pos_file_path = infinite_game.parameterized_agent_play(agent_parameters, self.map_name, False)

		action_string = trace_processor.actions_to_string_translator(action_list)

		#fitness = - trace_processor.lev_distance(action_string, cluster_example)

		fitness = - trace_processor.len_discounted_lev_distance(action_string, self.cluster_example)

		# print("Action String: ", action_string)
		# print("Cluster: ", cluster_example)
		# print("Fitness: ", fitness)

		return fitness, action_string, pos_file_path



	def get_next_generation(self):

		fitness_list = []
		summed_fitness = 0
		for agent in self.population:
			fitness, _, _ = self.get_fitness(agent)
			summed_fitness += fitness
			fitness_list.append(fitness)

		average_fitness = summed_fitness/len(self.population)

		fitness_ordered_agents = [x for _,x in sorted(zip(fitness_list, self.population), reverse=True)]

		print("Ordered Fitness: ", sorted(fitness_list, reverse=True))

		best_fitness = max(fitness_list)
		best_agent = fitness_ordered_agents[0]

		survivors = fitness_ordered_agents[:self.truncation_selection]
		children = []

		while((len(survivors) + len(children)) < self.population_size ):

			#choose 2 prents from the survivors
			parents = random.sample(survivors, 2)

			#create a child that is a mixture of both with a sprinkle of mutation
			child = []
			for parameter_num in range(len(parents[0])):
				if random.uniform(0, 1) < self.mutation_rate:
					parameter = random.randint(0,100)
				else:
					parameter = random.choice([parents[0][parameter_num], parents[1][parameter_num]])
				child.append(parameter)
			children.append(child)

		new_population = survivors + children

		return new_population, average_fitness, best_agent, best_fitness


		

	def run_evolution(self):

		self.get_initial_population()

		for i in range(max_generations):
			print("Generation ", i + 1 )
			self.population, average_fitness, best_agent, best_fitness = self.get_next_generation()
			self.average_fitness_evolution.append(average_fitness)
			self.best_fitness_evolution.append(best_fitness)
			self.best_agent_evolution.append(best_agent)
			gc.collect()


def run_evolution_exemplars(population_size, truncation_percentage, mutation_rate, max_generations, map_name):

	cluster_exemplars_list = glob.glob("./Saved_Clusters/" + map_name + "/*/cluster_exemplar.txt")

	print(cluster_exemplars_list)


	for cluster_file in cluster_exemplars_list:

		f = open(cluster_file)

		cluster_example_file = f.read()

		f.close()

		cluster_example = trace_processor.actions_to_string_translator(trace_processor.file_to_actions_translator(cluster_example_file))
		#We need to get them from the trace_processor file and chose the one we are interested in

		evo = Evo(population_size, truncation_percentage, mutation_rate, max_generations, cluster_example, map_name)

		evo.run_evolution()

		save_name = "./Saved_Personas/Persona_Evolution" + cluster_file.replace("\\", "/").replace("/cluster_exemplar", "").replace("./Saved_Clusters", "")

		file = open(save_name, 'a')

		file.write("\nFinal Population: " + str(evo.population))

		print("\n\n\nFinal Population: ", evo.population)

		file.write("\nAverage Fitness Evolution: " + str(evo.average_fitness_evolution))

		print("\n\n\nAverage Fitness Evolution: ", evo.average_fitness_evolution)

		file.write("\nBest Fitness Evolution: " + str(evo.best_fitness_evolution))

		print("\n\n\nBest Fitness Evolution: ", evo.best_fitness_evolution)

		file.write("\nBest Agent Evolution: " + str(evo.best_agent_evolution))

		print("\n\n\nBest Agent Evolution: ", evo.best_agent_evolution)

		#Save trace of best agent

		fitness, action_string, pos_file_path = evo.get_fitness(evo.best_agent_evolution[-1])

		save_name = "./Saved_Personas/Location_Traces" + cluster_file.replace("\\", "/").replace("/cluster_exemplar", "").replace("./Saved_Clusters", "").replace(".txt", "")

		predictor.save_location(pos_file_path, 1, save_name)


def run_evolution_all_traces(population_size, truncation_percentage, mutation_rate, max_generations):



	traces_list = glob.glob("./First_Study/*/Traces_Actions_*.txt")





	for trace_file in traces_list:


		cluster_example = trace_processor.actions_to_string_translator(trace_processor.file_to_actions_translator(trace_file))
		#We need to get them from the trace_processor file and chose the one we are interested in

		map_name = trace_file.split("Traces_Actions_")[1][:6]

		evo = Evo(population_size, truncation_percentage, mutation_rate, max_generations, cluster_example, map_name)

		evo.run_evolution()

		save_name = "./Saved_Personas/Persona_Evolution/" + trace_file.replace("\\", "/").split("/")[-1]

		file = open(save_name, 'a')

		file.write("\nFinal Population: " + str(evo.population))

		print("\n\n\nFinal Population: ", evo.population)

		file.write("\nAverage Fitness Evolution: " + str(evo.average_fitness_evolution))

		print("\n\n\nAverage Fitness Evolution: ", evo.average_fitness_evolution)

		file.write("\nBest Fitness Evolution: " + str(evo.best_fitness_evolution))

		print("\n\n\nBest Fitness Evolution: ", evo.best_fitness_evolution)

		file.write("\nBest Agent Evolution: " + str(evo.best_agent_evolution))

		print("\n\n\nBest Agent Evolution: ", evo.best_agent_evolution)

		#Save trace of best agent

		fitness, action_string, pos_file_path = evo.get_fitness(evo.best_agent_evolution[-1])

		save_name = "./Saved_Personas/Location_Traces/" + trace_file.replace("\\", "/").split("/")[-1].replace(".txt", "")

		predictor.save_location(pos_file_path, 1, save_name)



if __name__ == '__main__':


	population_size = 10
	truncation_percentage = 0.3
	mutation_rate = 0.2
	max_generations = 2

	map_name = "Level1"


	run_evolution_exemplars(population_size, truncation_percentage, mutation_rate, max_generations, map_name)

	#run_evolution_all_traces(population_size, truncation_percentage, mutation_rate, max_generations)