def evaluate_pong_performance(test_brain, visualization_mode):
#top_indices = [87, 79, 80, 77, 112, 1, 8, 72, 6, 28, 3, 110, 82, 85, 78, 9, 81, 90, 106, 74]
best_score = 0
desired_score = 2000.0
brain_speed = 5
trials = 100
output_count = 3
for i in range(trials):
env = gym.make('Pong-ram-v0')
test_instance = copy.deepcopy(test_brain)
observations = env.reset()
score = 0
while 1:
#score += 1
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
env.render()
output = [0] * output_count
#inputs = utils.extract_observations(top_indices, observations)
for i in range(len(observations)):
brain.Mutation_params().upper_input_bounds[i] = max(brain.Mutation_params().upper_input_bounds[i],observations[i])
brain.Mutation_params().lower_input_bounds[i] = min(brain.Mutation_params().lower_input_bounds[i],observations[i])
raw_output = []
for i in range(brain_speed):
raw_output.append ( test_instance.advance(observations, 3))
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
visualization.visualize_brain(brain.print_brain_to_json(test_instance))
for i in range(output_count):
for c in range(brain_speed):
output[i] += raw_output[c][i]
output[i] = int(output[i] > int(brain_speed/2))
action = min(utils.binary_array_to_decimal(output), 5)
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
print('ACTION: ' + str(action))
observations,reward,done,info = env.step(action)
score += 1#reward
if done:
best_score += score #+ 21
env.close()
break
return ((best_score)/(desired_score* trials)) * 100
def evaluate_chopper_performance(test_brain, visualization_mode):
#top_indices = [87, 79, 80, 77, 112, 1, 8, 72, 6, 28, 3, 110, 82, 85, 78, 9, 81, 90, 106, 74]
best_score = 0.0
desired_score = 1000
trials = 5
output_count = 5
brain_speed = 5
for i in range(trials):
env = gym.make('ChopperCommand-ram-v0')
observations = env.reset()
score = 0
for c in range(1000):
#score += 1
if visualization_mode == Learning_flags.VISUALIZATION_ON:
env.render()
output = [0] * 5
#inputs = utils.extract_observations(top_indices, observations)
for i in range(len(observations)):
brain.Mutation_params().upper_input_bounds[i] = max(
brain.Mutation_params().upper_input_bounds[i],
observations[i])
brain.Mutation_params().lower_input_bounds[i] = min(
brain.Mutation_params().lower_input_bounds[i],
observations[i])
raw_output = []
for i in range(brain_speed):
raw_output.append(test_brain.advance(observations, 5))
if visualization_mode == Learning_flags.VISUALIZATION_ON:
visualization.visualize_brain(
brain.print_brain_to_json(test_brain))
for i in range(output_count):
for c in range(brain_speed):
output[i] += raw_output[c][i]
output[i] = int(output[i] > int(output_count / 2))
action = min(utils.binary_array_to_decimal(output), 17)
if visualization_mode == Learning_flags.VISUALIZATION_ON:
print('ACTION: ' + str(action))
observations, reward, done, info = env.step(action)
score += reward
if done:
break
best_score += score
env.close()
return (best_score / (desired_score * trials)) * 100
def evalute_pendulum_cart_performance(test_brain, visualization_mode):
total_score = 0.0
desired_score = 200
trials = 100
for c in range(trials):
env = gym.make('CartPole-v0')
observations = env.reset()
test_instance = copy.deepcopy(test_brain)
score = 0
while True:#for h in range(desired_score):
#score += 1
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
env.render()
output = [0] * 3
for i in range(len(observations)):
brain.Mutation_params().upper_input_bounds[i] = max(brain.Mutation_params().upper_input_bounds[i],observations[i])
brain.Mutation_params().lower_input_bounds[i] = min(brain.Mutation_params().lower_input_bounds[i],observations[i])
sum = 0
for i in range(5):
sum += test_instance.advance(observations, 1)[0]
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
visualization.visualize_brain(brain.print_brain_to_json(test_instance))
action = int(sum >=3)
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
print('ACTION: ' + str(action))
observations,reward,done,info = env.step(action)
score += reward
if done:
total_score += score
break
env.close()
return (total_score / (trials * desired_score)) * 100
def evaluate_potion_store_performance(test_brain, visualization_mode):
time_limit = 200
time = 0
potion_count = 6
health_potion_count = 2
incorrect_limit = 5
correct_limit = 100.0
brain_speed = 5
total_score = 0
indices = [i for i in range(potion_count)]
coms = combinations(indices, health_potion_count)
for com in coms:
assert (health_potion_count <= potion_count)
test_instance = copy.deepcopy(test_brain)
health_potion_indices = com
correct_potions = 0.0
incorrect_potions = 0.0
correct_bit = 0
incorrect_bit = 0
time = 0
while correct_potions < correct_limit and incorrect_potions < incorrect_limit and time < time_limit:
potion_offer = randrange(potion_count)
input = [0] * potion_count
input[potion_offer] = 1
input = [correct_bit, incorrect_bit] + input
assert(len(input)== 8)
output = 0
for i in range(brain_speed):
output += test_instance.advance(input, 1)[0]
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
visualization.visualize_brain(brain.print_brain_to_json(test_instance))
output = int(output > int(brain_speed/2))
if output:
if (potion_offer not in health_potion_indices):
incorrect_potions += 1
incorrect_bit = 1
correct_bit = 0
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
print('DRANK ' + str(potion_offer) + ': POISION')
else:
correct_potions += 1
correct_bit = 1
incorrect_bit = 0
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
print('DRANK ' + str(potion_offer) + ': HEALTH POTION')
else:
correct_bit = 0
incorrect_bit = 0
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
print('DECLINED OFFER FOR ' + str(potion_offer))
time += 1
#if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
#test_instance.print_activation_record()
total_score += correct_potions
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
print("NUMBER CORRECT: " + str(correct_potions))
return float(total_score/(correct_limit*15)) * 100.0