agents.ipynb, MemoryError: In call to configurable 'WrappedReplayBuffer'

[lucifer2288]

I think the replay_capacity should be to 1000? but i could't find which variable control this parameter? INFO:tensorflow:Creating MyRandomDQNAgent agent with the following parameters: INFO:tensorflow: gamma: 0.990000 INFO:tensorflow: update_horizon: 1.000000 INFO:tensorflow: min_replay_history: 20000 INFO:tensorflow: update_period: 4 INFO:tensorflow: target_update_period: 8000 INFO:tensorflow: epsilon_train: 0.010000 INFO:tensorflow: epsilon_eval: 0.001000 INFO:tensorflow: epsilon_decay_period: 250000 INFO:tensorflow: tf_device: /cpu:* INFO:tensorflow: use_staging: True INFO:tensorflow: optimizer: <tensorflow.python.training.rmsprop.RMSPropOptimizer object at 0x0000000005C7ABE0> INFO:tensorflow:Creating a OutOfGraphReplayBuffer replay memory with the following parameters: INFO:tensorflow: observation_shape: (84, 84) INFO:tensorflow: stack_size: 4 INFO:tensorflow: replay_capacity: 1000000 INFO:tensorflow: batch_size: 32 INFO:tensorflow: update_horizon: 1 INFO:tensorflow: gamma: 0.990000

MemoryError Traceback (most recent call last) in 28 training_steps=10, 29 evaluation_steps=10, ---> 30 max_steps_per_episode=100)

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\config.py in wrapper(*args, **kwargs) 1030 scope_info = " in scope '{}'".format(scope_str) if scope_str else '' 1031 err_str = err_str.format(name, fn, scope_info) -> 1032 utils.augment_exception_message_and_reraise(e, err_str) 1033 1034 return wrapper

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\utils.py in augment_exception_message_and_reraise(exception, message) 46 if six.PY3: 47 ExceptionProxy.qualname = type(exception).qualname ---> 48 six.raise_from(proxy.with_traceback(exception.traceback), None) 49 else: 50 six.reraise(proxy, None, sys.exc_info()[2])

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\six.py in raise_from(value, from_value)

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\config.py in wrapper(*args, **kwargs) 1007 1008 try: -> 1009 return fn(*new_args, **new_kwargs) 1010 except Exception as e: # pylint: disable=broad-except 1011 err_str = ''

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\dopamine\atari\run_experiment.py in init(self, base_dir, create_agent_fn, create_environment_fn, game_name, sticky_actions, checkpoint_file_prefix, logging_file_prefix, log_every_n, num_iterations, training_steps, evaluation_steps, max_steps_per_episode) 164 config=tf.ConfigProto(allow_soft_placement=True)) 165 self._agent = create_agent_fn(self._sess, self._environment, --> 166 summary_writer=self._summary_writer) 167 self._summary_writer.add_graph(graph=tf.get_default_graph()) 168 self._sess.run(tf.global_variables_initializer())

in create_random_dqn_agent(sess, environment, summary_writer) 16 def create_random_dqn_agent(sess, environment, summary_writer=None): 17 """The Runner class will expect a function of this type to create an agent.""" ---> 18 return MyRandomDQNAgent(sess, num_actions=environment.action_space.n) 19 20 # Create the runner class with this agent. We use very small numbers of steps

in init(self, sess, num_actions) 6 def init(self, sess, num_actions): 7 """This maintains all the DQN default argument values.""" ----> 8 super(MyRandomDQNAgent, self).init(sess, num_actions) 9 10 def step(self, reward, observation):

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\config.py in wrapper(*args, **kwargs) 1030 scope_info = " in scope '{}'".format(scope_str) if scope_str else '' 1031 err_str = err_str.format(name, fn, scope_info) -> 1032 utils.augment_exception_message_and_reraise(e, err_str) 1033 1034 return wrapper

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\utils.py in augment_exception_message_and_reraise(exception, message) 46 if six.PY3: 47 ExceptionProxy.qualname = type(exception).qualname ---> 48 six.raise_from(proxy.with_traceback(exception.traceback), None) 49 else: 50 six.reraise(proxy, None, sys.exc_info()[2])

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\six.py in raise_from(value, from_value)

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\config.py in wrapper(*args, **kwargs) 1007 1008 try: -> 1009 return fn(*new_args, **new_kwargs) 1010 except Exception as e: # pylint: disable=broad-except 1011 err_str = ''

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\dopamine\agents\dqn\dqn_agent.py in init(self, sess, num_actions, observation_shape, observation_dtype, stack_size, gamma, update_horizon, min_replay_history, update_period, target_update_period, epsilon_fn, epsilon_train, epsilon_eval, epsilon_decay_period, tf_device, use_staging, max_tf_checkpoints_to_keep, optimizer, summary_writer, summary_writing_frequency) 174 self.state_ph = tf.placeholder(observation_dtype, state_shape, 175 name='state_ph') --> 176 self._replay = self._build_replay_buffer(use_staging) 177 178 self._build_networks()

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\dopamine\agents\dqn\dqn_agent.py in _build_replay_buffer(self, use_staging) 261 use_staging=use_staging, 262 update_horizon=self.update_horizon, --> 263 gamma=self.gamma) 264 265 def _build_target_q_op(self):

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\config.py in wrapper(*args, **kwargs) 1030 scope_info = " in scope '{}'".format(scope_str) if scope_str else '' 1031 err_str = err_str.format(name, fn, scope_info) -> 1032 utils.augment_exception_message_and_reraise(e, err_str) 1033 1034 return wrapper

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\utils.py in augment_exception_message_and_reraise(exception, message) 46 if six.PY3: 47 ExceptionProxy.qualname = type(exception).qualname ---> 48 six.raise_from(proxy.with_traceback(exception.traceback), None) 49 else: 50 six.reraise(proxy, None, sys.exc_info()[2])

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\six.py in raise_from(value, from_value)

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\gin\config.py in wrapper(*args, **kwargs) 1007 1008 try: -> 1009 return fn(*new_args, **new_kwargs) 1010 except Exception as e: # pylint: disable=broad-except 1011 err_str = ''

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\dopamine\replay_memory\circular_replay_buffer.py in init(self, observation_shape, stack_size, use_staging, replay_capacity, batch_size, update_horizon, gamma, wrapped_memory, max_sample_attempts, extra_storage_types, observation_dtype) 705 update_horizon, gamma, max_sample_attempts, 706 observation_dtype=observation_dtype, --> 707 extra_storage_types=extra_storage_types) 708 709 self.create_sampling_ops(use_staging)

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\dopamine\replay_memory\circular_replay_buffer.py in init(self, observation_shape, stack_size, replay_capacity, batch_size, update_horizon, gamma, max_sample_attempts, extra_storage_types, observation_dtype) 156 else: 157 self._extra_storage_types = [] --> 158 self._create_storage() 159 self.add_count = np.array(0) 160 self.invalid_range = np.zeros((self._stack_size))

c:\users\gyang\my_prodect\dopamine-env\lib\site-packages\dopamine\replay_memory\circular_replay_buffer.py in _create_storage(self) 172 array_shape = [self._replay_capacity] + list(storage_element.shape) 173 self._store[storage_element.name] = np.empty( --> 174 array_shape, dtype=storage_element.type) 175 176 def get_add_args_signature(self):

MemoryError: In call to configurable 'WrappedReplayBuffer' (<function WrappedReplayBuffer.init at 0x000000001441E488>) In call to configurable 'DQNAgent' (<function DQNAgent.init at 0x0000000012000AE8>) In call to configurable 'Runner' (<function Runner.init at 0x0000000016573488>)

[yakotaki]

https://github.com/google/dopamine/blob/bc66570375dd8d4ad785948e032a7662bb32e2a6/dopamine/replay_memory/circular_replay_buffer.py#L653

[mymusise]

Try to override the config before define Runner with:

gin_bindings = [
    'WrappedReplayBuffer.replay_capacity = 100'  # To prevent OOM.
]
run_experiment.load_gin_configs([], gin_bindings)  # Overwrite it.

random_dqn_runner = run_experiment.Runner(...)