def __init__(self, config=config):
# Reproduce training results
np.random.seed(42)
tf.random.set_seed(42)
self._config = config
current_dir = os.getcwd()
model_save_path = current_dir + '.\Model'
# Apply env wrappers
env = gym.make(config['Environment Name'])
env = preprocess.FrameSkip(env, 4)
env = preprocess.GrayScaleImage(env,
height=84,
width=84,
grayscale=True)
env = preprocess.FrameStack(env, 4)
NUM_STATE = env.observation_space.shape
NUM_ACTIONS = env.env.action_space.n
ACTION_SPACE = env.env.action_space
network_params = (NUM_STATE, 1.0, NUM_ACTIONS, ACTION_SPACE)
self.Test_Model = None
if self._config['CNN type'] == 'large':
self.Test_Model = AC_Model_Large(NUM_STATE,
NUM_ACTIONS,
self._config,
is_training=False)
else:
self.Test_Model = AC_Model_Small(NUM_STATE,
NUM_ACTIONS,
self._config,
is_training=False)
# Load model if exists
if not os.path.exists(model_save_path):
os.makedirs(model_save_path)
else:
try:
if (os.path.exists(model_save_path + "\checkpoint")):
self.Test_Model.load_model_weights()
print("Model restored.")
print("Now running environment...")
self._run(10000, env, self.Test_Model, True)
else:
raise ("There is no model available")
except:
print("ERROR: There was an issue running the model!")
raise
def __init__(self,
render=False,
eps_start=EPS_START,
eps_end=EPS_STOP,
eps_steps=EPS_STEPS):
threading.Thread.__init__(self)
self.render = render
# Make the super mario gym environment and apply wrappers
self.env = gym.make(ENV)
self.env = BinarySpaceToDiscreteSpaceEnv(self.env, SIMPLE_MOVEMENT)
self.env = preprocess.GrayScaleImage(self.env,
height=HIGHT,
width=WIDTH,
grayscale=True)
# self.env = wrappers.Monitor(self.env, "./Super_Mario_AI/videos", force = True, write_upon_reset=True)
self.agent = Agent(TEMPERATURE)
###### Main ######
##################
### Here is a list of the available environments ###
# SuperMarioBros-v0, SuperMarioBros-v1, SuperMarioBros-v2,
# SuperMarioBros-v3, SuperMarioBrosNoFrameskip-v0, SuperMarioBrosNoFrameskip-v1,
# SuperMarioBrosNoFrameskip-v2, SuperMarioBrosNoFrameskip-v3, SuperMarioBros2-v0,
# SuperMarioBros2-v1, SuperMarioBros2NoFrameskip-v0, SuperMarioBros2NoFrameskip-v1
# Get and build our test environment
env = gym_super_mario_bros.make('SuperMarioBros-v0')
# env = BinarySpaceToDiscreteSpaceEnv(env, COMPLEX_MOVEMENT)
env = BinarySpaceToDiscreteSpaceEnv(env, SIMPLE_MOVEMENT)
env = preprocess.GrayScaleImage(env, height=128, width=128, grayscale=True)
env = wrappers.Monitor(env,
"./Super_Mario_AI/videos",
force=True,
write_upon_reset=True)
number_actions = env.action_space.n
# Construct/Load a model
cnn = CNN(number_actions)
optimizer = optim.Adam(cnn.parameters(), lr=0.001)
loss = nn.MSELoss()
last_epoch = 1
# If there is a previous save
if os.path.exists("./Super_Mario_AI/Model/model.pth"):
checkpoint = torch.load("./Super_Mario_AI/Model/model.pth")
def __init__(self, config):
self._config = config
def get_available_gpus():
local_device_protos = device_lib.list_local_devices()
return [
x.name for x in local_device_protos if x.device_type == 'GPU'
]
print("GPU Available: ", tf.test.is_gpu_available())
# GPU configuration
gpus = tf.config.experimental.list_physical_devices('GPU')
tf.config.threading.set_inter_op_parallelism_threads(0)
tf.config.threading.set_intra_op_parallelism_threads(0)
if gpus:
try:
# Currently, memory growth needs to be the same across GPUs
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
except RuntimeError as e:
print(e)
env_names = []
for _ in range(self._config['Number of worker threads']):
env_names.append(self._config['Environment Name'])
# Configuration
# current_dir = os.getcwd()
self._model_save_path = '.\Model'
self._video_save_path = '.\Videos'
self.record = True
# Make the super mario gym environments and apply wrappers
self._envs = []
collector = Collector()
collector.set_dimensions(
["CMA", "EMA", "SMA", "LENGTH", "LOSS", 'TOTAL_EPISODE_REWARDS'])
self._plot = AsynchronousPlot(collector, live=False)
# Apply env wrappers
counter = 0
for env_name in env_names:
env = gym.make(env_name)
if env_name == 'SuperMarioBros-v0':
env = JoypadSpace(env, COMPLEX_MOVEMENT)
# Load wrapper class
env = Stats(env, collector)
if self._config['Wrapper class'] != '':
env = env_wrapper_import(self._config['Wrapper class'], env)
env = Monitor(env,
env.observation_space.shape,
savePath=self._video_save_path,
record=self.record)
env = preprocess.GrayScaleImage(
env, height=84, width=84, grayscale=self._config['Grayscale'])
env = preprocess.FrameStack(env, 4)
self._envs.append(env)
self.NUM_STATE = self._envs[0].observation_space.shape
self.NUM_ACTIONS = self._envs[0].env.action_space.n
self.ACTION_SPACE = self._envs[0].env.action_space
if not os.path.exists(self._video_save_path):
os.makedirs(self._video_save_path)
if not os.path.exists(self._model_save_path):
os.makedirs(self._model_save_path)
if not os.path.exists('.\stats'):
os.makedirs('.\stats')