def test_binary_array_to_decimal(self):
input = [1,0,1,0,0,1]
expected_output = 41
self.assertEqual(utils.binary_array_to_decimal(input), expected_output)
input = [1,0,1,0,1,0]
expected_output = 42
self.assertEqual(utils.binary_array_to_decimal(input), expected_output)
input = [1,1,0,0,1,1]
expected_output = 51
self.assertEqual(utils.binary_array_to_decimal(input), expected_output)
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 advance_agents(self, visualization_mode):
## get randomly ordered list of agents, sense, run, harvest actuation and publish to action_queue
agent_keys = list(self.agents.keys())
shuffle(agent_keys)
for key in agent_keys: #useful to note here that each 'key' is a tuple containing agent location in (x,y) format
agent = self.agents[key]
assert (self.grid[key[1]][key[0]] == Object_type.AGENT)
agent.energy -= 1
#sense
observations = self.sense(key, agent.direction)
output = []
for i in range(
len(observations)
): ## setting our bounds appropriately for threshold mutations
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])
result = agent.brain.advance_n_with_mode(
observations, brain.Mutation_params.output_count, 10,
visualization_mode)
numerical_result = utils.binary_array_to_decimal(result)
# print(result)
# print(numerical_result)
if visualization_mode == visualization.Visualization_flags.VISUALIZATION_ON:
print(result)
print(numerical_result)
agent.generate_action(numerical_result, self, key)
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 evaluate_space_invaders_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 = 500
trials = 100
for i in range(trials):
env = gym.make('SpaceInvaders-ram-v0')
observations = env.reset()
score = 0
while 1:
#score += 1
if visualization_mode == Learning_flags.VISUALIZATION_ON:
env.render()
output = [0] * 3
#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])
for i in range(1):
output = test_brain.advance(observations, 3)
action = min(utils.binary_array_to_decimal(output), 5)
if visualization_mode == Learning_flags.VISUALIZATION_ON:
print('ACTION: ' + str(action))
observations, reward, done, info = env.step(action)
score += reward
if done:
best_score += score
env.close()
break
return (best_score / (desired_score * trials)) * 100