def __init__(self, sess, env, brain_name, trainer_parameters, training, seed):
"""
Responsible for collecting experiences and training PPO model.
:param sess: Tensorflow session.
:param env: The UnityEnvironment.
:param trainer_parameters: The parameters for the trainer (dictionary).
:param training: Whether the trainer is set for training.
"""
self.param_keys = ['brain_to_imitate', 'batch_size', 'time_horizon', 'graph_scope',
'summary_freq', 'max_steps', 'batches_per_epoch', 'use_recurrent', 'hidden_units',
'num_layers', 'sequence_length', 'memory_size']
for k in self.param_keys:
if k not in trainer_parameters:
raise UnityTrainerException("The hyperparameter {0} could not be found for the Imitation trainer of "
"brain {1}.".format(k, brain_name))
super(BehavioralCloningTrainer, self).__init__(sess, env, brain_name, trainer_parameters, training)
self.variable_scope = trainer_parameters['graph_scope']
self.brain_to_imitate = trainer_parameters['brain_to_imitate']
self.batches_per_epoch = trainer_parameters['batches_per_epoch']
self.use_recurrent = trainer_parameters['use_recurrent']
self.step = 0
self.sequence_length = 1
self.m_size = None
if self.use_recurrent:
self.m_size = trainer_parameters["memory_size"]
self.sequence_length = trainer_parameters["sequence_length"]
self.n_sequences = max(int(trainer_parameters['batch_size'] / self.sequence_length), 1)
self.cumulative_rewards = {}
self.episode_steps = {}
self.stats = {'losses': [], 'episode_length': [], 'cumulative_reward': []}
self.training_buffer = Buffer()
self.is_continuous_action = (env.brains[brain_name].vector_action_space_type == "continuous")
self.is_continuous_observation = (env.brains[brain_name].vector_observation_space_type == "continuous")
self.use_observations = (env.brains[brain_name].number_visual_observations > 0)
if self.use_observations:
logger.info('Cannot use observations with imitation learning')
self.use_states = (env.brains[brain_name].vector_observation_space_size > 0)
self.summary_path = trainer_parameters['summary_path']
if not os.path.exists(self.summary_path):
os.makedirs(self.summary_path)
self.summary_writer = tf.summary.FileWriter(self.summary_path)
with tf.variable_scope(self.variable_scope):
tf.set_random_seed(seed)
self.model = BehavioralCloningModel(
h_size=int(trainer_parameters['hidden_units']),
lr=float(trainer_parameters['learning_rate']),
n_layers=int(trainer_parameters['num_layers']),
m_size=self.m_size,
normalize=False,
use_recurrent=trainer_parameters['use_recurrent'],
brain=self.brain)
def test_dc_bc_model():
d_action_c_state_start = '''{
"AcademyName": "RealFakeAcademy",
"resetParameters": {},
"brainNames": ["RealFakeBrain"],
"externalBrainNames": ["RealFakeBrain"],
"logPath":"RealFakePath",
"apiNumber":"API-3",
"brainParameters": [{
"vectorObservationSize": 3,
"numStackedVectorObservations": 2,
"vectorActionSize": 2,
"memorySize": 0,
"cameraResolutions": [{"width":30,"height":40,"blackAndWhite":false}],
"vectorActionDescriptions": ["",""],
"vectorActionSpaceType": 0,
"vectorObservationSpaceType": 1
}]
}'''.encode()
tf.reset_default_graph()
with mock.patch('subprocess.Popen'):
with mock.patch('socket.socket') as mock_socket:
with mock.patch('glob.glob') as mock_glob:
with tf.Session() as sess:
with tf.variable_scope("FakeGraphScope"):
mock_glob.return_value = ['FakeLaunchPath']
mock_socket.return_value.accept.return_value = (
mock_socket, 0)
mock_socket.recv.return_value.decode.return_value = d_action_c_state_start
env = UnityEnvironment(' ')
model = BehavioralCloningModel(
env.brains["RealFakeBrain"])
init = tf.global_variables_initializer()
sess.run(init)
run_list = [model.sample_action, model.policy]
feed_dict = {
model.batch_size:
2,
model.dropout_rate:
1.0,
model.sequence_length:
1,
model.vector_in:
np.array([[1, 2, 3, 1, 2, 3], [3, 4, 5, 3, 4, 5]]),
model.visual_in[0]:
np.ones([2, 40, 30, 3])
}
sess.run(run_list, feed_dict=feed_dict)
env.close()
def test_cc_bc_model(mock_communicator, mock_launcher):
tf.reset_default_graph()
with tf.Session() as sess:
with tf.variable_scope("FakeGraphScope"):
mock_communicator.return_value = MockCommunicator(
discrete_action=False, visual_inputs=0)
env = UnityEnvironment(' ')
model = BehavioralCloningModel(env.brains["RealFakeBrain"])
init = tf.global_variables_initializer()
sess.run(init)
run_list = [model.sample_action, model.policy]
feed_dict = {model.batch_size: 2,
model.sequence_length: 1,
model.vector_in: np.array([[1, 2, 3, 1, 2, 3],
[3, 4, 5, 3, 4, 5]])}
sess.run(run_list, feed_dict=feed_dict)
env.close()