def _setup_graph(self, ob_space, ac_space):
self.x = tf.placeholder(tf.float32, [None] + list(ob_space.shape))
dist_class, self.logit_dim = ModelCatalog.get_action_dist(
ac_space, self.config["model"])
self._model = LSTM(self.x, self.logit_dim, {})
self.state_in = self._model.state_in
self.state_out = self._model.state_out
self.logits = self._model.outputs
self.action_dist = dist_class(self.logits)
# with tf.variable_scope("vf"):
# vf_model = ModelCatalog.get_model(self.x, 1)
self.vf = tf.reshape(
linear(self._model.last_layer, 1, "value", normc_initializer(1.0)),
[-1])
self.sample = self.action_dist.sample()
self.var_list = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
tf.get_variable_scope().name)
self.global_step = tf.get_variable("global_step", [],
tf.int32,
initializer=tf.constant_initializer(
0, dtype=tf.int32),
trainable=False)
def get_model(inputs,
num_outputs,
options=None,
state_in=None,
seq_lens=None):
"""Returns a suitable model conforming to given input and output specs.
Args:
inputs (Tensor): The input tensor to the model.
num_outputs (int): The size of the output vector of the model.
options (dict): Optional args to pass to the model constructor.
state_in (list): Optional RNN state in tensors.
seq_in (Tensor): Optional RNN sequence length tensor.
Returns:
model (Model): Neural network model.
"""
options = options or {}
model = ModelCatalog._get_model(inputs, num_outputs, options, state_in,
seq_lens)
if options.get("use_lstm"):
model = LSTM(model.last_layer, num_outputs, options, state_in,
seq_lens)
return model
def get_model(input_dict,
obs_space,
num_outputs,
options,
state_in=None,
seq_lens=None):
"""Returns a suitable model conforming to given input and output specs.
Args:
input_dict (dict): Dict of input tensors to the model, including
the observation under the "obs" key.
obs_space (Space): Observation space of the target gym env.
num_outputs (int): The size of the output vector of the model.
options (dict): Optional args to pass to the model constructor.
state_in (list): Optional RNN state in tensors.
seq_in (Tensor): Optional RNN sequence length tensor.
Returns:
model (Model): Neural network model.
"""
assert isinstance(input_dict, dict)
options = options or MODEL_DEFAULTS
model = ModelCatalog._get_model(input_dict, obs_space, num_outputs,
options, state_in, seq_lens)
if options.get("use_lstm"):
copy = dict(input_dict)
copy["obs"] = model.last_layer
model = LSTM(copy, obs_space, num_outputs, options, state_in,
seq_lens)
return model
def get_model(input_dict,
obs_space,
action_space,
num_outputs,
options,
state_in=None,
seq_lens=None):
"""Returns a suitable model conforming to given input and output specs.
Args:
input_dict (dict): Dict of input tensors to the model, including
the observation under the "obs" key.
obs_space (Space): Observation space of the target gym env.
action_space (Space): Action space of the target gym env.
num_outputs (int): The size of the output vector of the model.
options (dict): Optional args to pass to the model constructor.
state_in (list): Optional RNN state in tensors.
seq_lens (Tensor): Optional RNN sequence length tensor.
Returns:
model (models.Model): Neural network model.
"""
assert isinstance(input_dict, dict)
options = options or MODEL_DEFAULTS
model = ModelCatalog._get_model(input_dict, obs_space, action_space,
num_outputs, options, state_in,
seq_lens)
if options.get("use_lstm"):
copy = dict(input_dict)
copy["obs"] = model.last_layer
feature_space = gym.spaces.Box(-1,
1,
shape=(model.last_layer.shape[1], ))
model = LSTM(copy, feature_space, action_space, num_outputs,
options, state_in, seq_lens)
logger.debug(
"Created model {}: ({} of {}, {}, {}, {}) -> {}, {}".format(
model, input_dict, obs_space, action_space, state_in, seq_lens,
model.outputs, model.state_out))
model._validate_output_shape()
return model
def get_model(input_dict,
obs_space,
action_space,
num_outputs,
options,
state_in=None,
seq_lens=None):
"""Returns a suitable model conforming to given input and output specs.
Args:
input_dict (dict): Dict of input tensors to the model, including
the observation under the "obs" key.
obs_space (Space): Observation space of the target gym env.
action_space (Space): Action space of the target gym env.
num_outputs (int): The size of the output vector of the model.
options (dict): Optional args to pass to the model constructor.
state_in (list): Optional RNN state in tensors.
seq_lens (Tensor): Optional RNN sequence length tensor.
Returns:
model (models.Model): Neural network model.
"""
assert isinstance(input_dict, dict)
options = options or MODEL_DEFAULTS
model = ModelCatalog._get_model(input_dict, obs_space, action_space,
num_outputs, options, state_in,
seq_lens)
if options.get("use_lstm"):
copy = dict(input_dict)
copy["obs"] = model.last_layer
feature_space = gym.spaces.Box(
-1, 1, shape=(model.last_layer.shape[1], ))
model = LSTM(copy, feature_space, action_space, num_outputs,
options, state_in, seq_lens)
logger.debug(
"Created model {}: ({} of {}, {}, {}, {}) -> {}, {}".format(
model, input_dict, obs_space, action_space, state_in, seq_lens,
model.outputs, model.state_out))
model._validate_output_shape()
return model
