def evaluation(best, N, test_runs, output_dir, random=False):
print 'TEST RUN'
results = []
position = []
for _ in range(test_runs):
agents = [Agent([-1 for _ in range(IND_SIZE)]) for _ in range(NUM_IND)]
if random:
best = Agent([-1] * 32)
agents[0] = best
game = Game(agents, N)
game.play()
results.append(best.score)
position.append(
sorted([a.score for a in agents], reverse=True).index(best.score))
# Save results
if not os.path.exists(output_dir):
os.mkdir(output_dir)
plt.figure()
plt.hist(results)
plt.xlabel('Score')
plt.ylabel('Frequency')
plt.savefig(os.path.join(output_dir, 'scores.pdf'))
plt.figure()
plt.hist(position)
plt.xlabel('Rank')
plt.ylabel('Frequency')
plt.savefig(os.path.join(output_dir, 'ranks.pdf'))
with open(os.path.join(output_dir, 'scores.csv'), 'w') as f:
w = writer(f)
w.writerow(results)
with open(os.path.join(output_dir, 'ranks.csv'), 'w') as f:
w = writer(f)
w.writerow(position)
with open(os.path.join(output_dir, 'strategy.txt'), 'w') as f:
f.write(str(best.strategy))
return results, position
def singleAgentEvolution(CXPB, MUTPB, NGEN, N):
pop = toolbox.population(n=NUM_IND)
agents = [Agent(ind) for ind in pop]
game = Game(agents, N)
# Evaluate the entire population
fitnesses = game.play()
for ind, fit in zip(pop, fitnesses):
ind.fitness.values = fit
for g in range(NGEN):
# Select individual to be evolved
index = random.randint(0, len(pop) - 1)
# Select the new generation individuals
offspring = toolbox.select(pop, len(pop))
# Clone the selected individuals
offspring = map(toolbox.clone, offspring)
# Apply crossover and mutation on the offspring
child1 = offspring[index]
for child2 in [
element for idx, element in enumerate(pop) if idx != index
]:
if random.random() < CXPB:
toolbox.mate(child1, child2)
continue
mutant = offspring[index]
if random.random() < MUTPB:
toolbox.mutate(mutant)
# Set new (evolved) genotype
agents[index].setStrategy(pop[index])
# The chosen individual is replaced by the offspring
pop[index] = offspring[index]
# Evaluate the individuals
fitnesses = game.play()
for ind, fit in zip(pop, fitnesses):
ind.fitness.values = fit
# get best agent
best = agents[0]
for agent in agents:
if best.score < agent.score:
best = agent
return best
def multiAgentEvolution(CXPB, MUTPB, NGEN, N):
pop = toolbox.population(n=NUM_IND)
agents = [Agent(ind) for ind in pop]
game = Game(agents, N)
# Evaluate the entire population
fitnesses = game.play()
for ind, fit in zip(pop, fitnesses):
ind.fitness.values = fit
for g in range(NGEN):
# Select the next generation individuals
offspring = toolbox.select(pop, len(pop))
# Clone the selected individuals
offspring = map(toolbox.clone, offspring)
# Apply crossover and mutation on the offspring
for child1, child2 in zip(offspring[::2], offspring[1::2]):
if random.random() < CXPB:
toolbox.mate(child1, child2)
for mutant in offspring:
if random.random() < MUTPB:
toolbox.mutate(mutant)
# Set new (evolved) genotypes
for agent, ind in zip(agents, offspring):
agent.setStrategy(ind)
# Evaluate the individuals
fitnesses = game.play()
for ind, fit in zip(offspring, fitnesses):
ind.fitness.values = fit
# The population is entirely replaced by the offspring
pop[:] = offspring
# get best agent
best = agents[0]
for agent in agents:
if best.score < agent.score:
best = agent
return best
with open(os.path.join(output_dir, 'ranks.csv'), 'w') as f:
w = writer(f)
w.writerow(position)
with open(os.path.join(output_dir, 'strategy.txt'), 'w') as f:
f.write(str(best.strategy))
return results, position
if __name__ == '__main__':
CXPB, MUTPB, NGEN, N, TEST_RUNS = 0.5, 0.2, 10000, 1000, 1000
NUM_IND = 20
# Baseline agent who never changes society
best = Agent([-1] * 32)
evaluation(best, N, TEST_RUNS, 'baseline_20', random=True)
# All agents evolving at the same time
best = multiAgentEvolution(CXPB, MUTPB, NGEN, N)
evaluation(best, N, TEST_RUNS, 'multi_evolution_20')
# One agent evolving at a time
best = singleAgentEvolution(CXPB, MUTPB, NGEN, N)
evaluation(best, N, TEST_RUNS, 'single_evolution_20')
NUM_IND = 100
# Baseline agent who never changes society
best = Agent([-1] * 32)
evaluation(best, N, TEST_RUNS, 'baseline_100', random=True)
def __init__(self,sys1 = Agent(),sys2 = Agent()):
self.system_1_model = System1Agent()
self.system_2_model = System2Agent()
self.count = 0
def game_setup(num_agents):
game = Game()
agents = [Agent() for i in range(num_agents)]
ratings = [Rating() for i in range(num_agents)]
return game, agents, ratings
