code
stringlengths
66
870k
docstring
stringlengths
19
26.7k
func_name
stringlengths
1
138
language
stringclasses
1 value
repo
stringlengths
7
68
path
stringlengths
5
324
url
stringlengths
46
389
license
stringclasses
7 values
def __call__(self, value: typing.Any) -> trajectory.Transition: """Converts `value` to a Transition. Performs data validation and pruning. - If `value` is already a `Transition`, only validation is performed. - If `value` is a `Trajectory` and `squeeze_time_dim = True` then `value` it must have tenso...
Converts `value` to a Transition. Performs data validation and pruning. - If `value` is already a `Transition`, only validation is performed. - If `value` is a `Trajectory` and `squeeze_time_dim = True` then `value` it must have tensors with shape `[B, T=2]` outer dims. This is converted to a `Tran...
__call__
python
tensorflow/agents
tf_agents/agents/data_converter.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/data_converter.py
Apache-2.0
def __call__(self, value: typing.Any) -> trajectory.Transition: """Convert `value` to an N-step Transition; validate data & prune. - If `value` is already a `Transition`, only validation is performed. - If `value` is a `Trajectory` with tensors containing a time dimension having `T != n + 1`, a `Valu...
Convert `value` to an N-step Transition; validate data & prune. - If `value` is already a `Transition`, only validation is performed. - If `value` is a `Trajectory` with tensors containing a time dimension having `T != n + 1`, a `ValueError` is raised. Args: value: A `Trajectory` or `Transitio...
__call__
python
tensorflow/agents
tf_agents/agents/data_converter.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/data_converter.py
Apache-2.0
def __init__( self, data_context: DataContext, gamma: types.Float, n: typing.Optional[int] = None, ): """Create the AsNStepTransition converter. For more details on how `Trajectory` objects are converted to N-step `Transition` objects, see `tf_agents.trajectories.trajectory.to...
Create the AsNStepTransition converter. For more details on how `Trajectory` objects are converted to N-step `Transition` objects, see `tf_agents.trajectories.trajectory.to_n_step_transition`. Args: data_context: An instance of `DataContext`, typically accessed from the `TFAgent.data_con...
__init__
python
tensorflow/agents
tf_agents/agents/data_converter.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/data_converter.py
Apache-2.0
def __call__(self, value: typing.Any) -> trajectory.Transition: """Convert `value` to an N-step Transition; validate data & prune. - If `value` is already a `Transition`, only validation is performed. - If `value` is a `Trajectory` with tensors containing a time dimension having `T != n + 1`, a `Valu...
Convert `value` to an N-step Transition; validate data & prune. - If `value` is already a `Transition`, only validation is performed. - If `value` is a `Trajectory` with tensors containing a time dimension having `T != n + 1`, a `ValueError` is raised. Args: value: A `Trajectory` or `Transitio...
__call__
python
tensorflow/agents
tf_agents/agents/data_converter.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/data_converter.py
Apache-2.0
def test_loss_and_train_output( test: test_utils.TestCase, expect_equal_loss_values: bool, agent: tf_agent.TFAgent, experience: types.NestedTensor, weights: Optional[types.Tensor] = None, **kwargs ): """Tests that loss() and train() outputs are equivalent. Checks that the outputs have the s...
Tests that loss() and train() outputs are equivalent. Checks that the outputs have the same structures and shapes, and compares loss values based on `expect_equal_loss_values`. Args: test: An instance of `test_utils.TestCase`. expect_equal_loss_values: Whether to expect `LossInfo.loss` to have the same ...
test_loss_and_train_output
python
tensorflow/agents
tf_agents/agents/test_util.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/test_util.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, policy: tf_policy.TFPolicy, collect_policy: tf_policy.TFPolicy, train_sequence_length: Optional[int], num_outer_dims: int = 2, training_data_spec: Optional[types.NestedTensorSpec] = None...
Meant to be called by subclass constructors. Args: time_step_spec: A nest of tf.TypeSpec representing the time_steps. Provided by the user. action_spec: A nest of BoundedTensorSpec representing the actions. Provided by the user. policy: An instance of `tf_policy.TFPolicy` represen...
__init__
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def initialize(self) -> Optional[tf.Operation]: """Initializes the agent. Returns: An operation that can be used to initialize the agent. Raises: RuntimeError: If the class was not initialized properly (`super.__init__` was not called). """ if self._enable_functions and getattr...
Initializes the agent. Returns: An operation that can be used to initialize the agent. Raises: RuntimeError: If the class was not initialized properly (`super.__init__` was not called).
initialize
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def preprocess_sequence( self, experience: types.NestedTensor ) -> types.NestedTensor: """Defines preprocess_sequence function to be fed into replay buffers. This defines how we preprocess the collected data before training. Defaults to pass through for most agents. Structure of `experience` mu...
Defines preprocess_sequence function to be fed into replay buffers. This defines how we preprocess the collected data before training. Defaults to pass through for most agents. Structure of `experience` must match that of `self.collect_data_spec`. Args: experience: a `Trajectory` shaped [batch, ...
preprocess_sequence
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def train( self, experience: types.NestedTensor, weights: Optional[types.Tensor] = None, **kwargs ) -> LossInfo: """Trains the agent. Args: experience: A batch of experience data in the form of a `Trajectory`. The structure of `experience` must match that of `self.traini...
Trains the agent. Args: experience: A batch of experience data in the form of a `Trajectory`. The structure of `experience` must match that of `self.training_data_spec`. All tensors in `experience` must be shaped `[batch, time, ...]` where `time` must be equal to `self.train_step_leng...
train
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def loss( self, experience: types.NestedTensor, weights: Optional[types.Tensor] = None, training: bool = False, **kwargs ) -> LossInfo: """Gets loss from the agent. If the user calls this from _train, it must be in a `tf.GradientTape` scope in order to apply gradients to tra...
Gets loss from the agent. If the user calls this from _train, it must be in a `tf.GradientTape` scope in order to apply gradients to trainable variables. If intermediate gradient steps are needed, _loss and _train will return different values since _loss only supports updating all gradients at once ...
loss
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def training_data_spec(self) -> types.NestedTensorSpec: """Returns a trajectory spec, as expected by the train() function.""" if self._training_data_spec is not None: return self._training_data_spec else: return self.collect_data_spec
Returns a trajectory spec, as expected by the train() function.
training_data_spec
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def _preprocess_sequence( self, experience: types.NestedTensor ) -> types.NestedTensor: """Defines preprocess_sequence function to be fed into replay buffers. This defines how we preprocess the collected data before training. Defaults to pass through for most agents. Subclasses may override this. ...
Defines preprocess_sequence function to be fed into replay buffers. This defines how we preprocess the collected data before training. Defaults to pass through for most agents. Subclasses may override this. Args: experience: a `Trajectory` shaped [batch, time, ...] or [time, ...] which repre...
_preprocess_sequence
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def _loss( self, experience: types.NestedTensor, weights: types.Tensor, training: bool, **kwargs ) -> Optional[LossInfo]: """Computes loss. This method does not increment self.train_step_counter or upgrade gradients. By default, any networks are called with `training=False`....
Computes loss. This method does not increment self.train_step_counter or upgrade gradients. By default, any networks are called with `training=False`. Args: experience: A batch of experience data in the form of a `Trajectory`. The structure of `experience` must match that of `self.training_d...
_loss
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def _train( self, experience: types.NestedTensor, weights: types.Tensor ) -> LossInfo: """Returns an op to train the agent. This method *must* increment self.train_step_counter exactly once. TODO(b/126271669): Consider automatically incrementing this. Args: experience: A batch of experie...
Returns an op to train the agent. This method *must* increment self.train_step_counter exactly once. TODO(b/126271669): Consider automatically incrementing this. Args: experience: A batch of experience data in the form of a `Trajectory`. The structure of `experience` must match that of `self...
_train
python
tensorflow/agents
tf_agents/agents/tf_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/tf_agent.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, cloning_network: network.Network, optimizer: types.Optimizer, num_outer_dims: Literal[1, 2] = 1, # pylint: disable=bad-whitespace epsilon_greedy: types.Float = 0.1, loss_fn: Optional[ ...
Creates an instance of a Behavioral Cloning agent. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of BoundedTensorSpec representing the actions. cloning_network: A `tf_agents.networks.Network` to be used by the agent. The network will be called as ...
__init__
python
tensorflow/agents
tf_agents/agents/behavioral_cloning/behavioral_cloning_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/behavioral_cloning/behavioral_cloning_agent.py
Apache-2.0
def create_arbitrary_trajectory(): """Creates an arbitrary trajectory for unit testing BehavioralCloningAgent. This trajectory contains Tensors shaped `[6, 1, ...]` where `6` is the number of time steps and `1` is the batch. Observations are unbounded but actions are bounded to take values within `[1, 2]`. ...
Creates an arbitrary trajectory for unit testing BehavioralCloningAgent. This trajectory contains Tensors shaped `[6, 1, ...]` where `6` is the number of time steps and `1` is the batch. Observations are unbounded but actions are bounded to take values within `[1, 2]`. The action space is discrete. Policy ...
create_arbitrary_trajectory
python
tensorflow/agents
tf_agents/agents/behavioral_cloning/behavioral_cloning_agent_test.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/behavioral_cloning/behavioral_cloning_agent_test.py
Apache-2.0
def verifyTrainAndRestore( self, observation_spec, action_spec, actor_net, loss_fn=None ): """Helper function for testing correct variable updating and restoring.""" batch_size = 2 observations = tensor_spec.sample_spec_nest( observation_spec, outer_dims=(batch_size,) ) actions = ten...
Helper function for testing correct variable updating and restoring.
verifyTrainAndRestore
python
tensorflow/agents
tf_agents/agents/behavioral_cloning/behavioral_cloning_agent_test.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/behavioral_cloning/behavioral_cloning_agent_test.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, categorical_q_network: network.Network, optimizer: types.Optimizer, observation_and_action_constraint_splitter: Optional[ types.Splitter ] = None, min_q_value: types.Float = -1...
Creates a Categorical DQN Agent. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A `BoundedTensorSpec` representing the actions. categorical_q_network: A categorical_q_network.CategoricalQNetwork that returns the q_distribution for each action. optim...
__init__
python
tensorflow/agents
tf_agents/agents/categorical_dqn/categorical_dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/categorical_dqn_agent.py
Apache-2.0
def _loss( self, experience, td_errors_loss_fn=tf.compat.v1.losses.huber_loss, gamma=1.0, reward_scale_factor=1.0, weights=None, training=False, ): """Computes critic loss for CategoricalDQN training. See Algorithm 1 and the discussion immediately preceding it in pag...
Computes critic loss for CategoricalDQN training. See Algorithm 1 and the discussion immediately preceding it in page 6 of "A Distributional Perspective on Reinforcement Learning" Bellemare et al., 2017 https://arxiv.org/abs/1707.06887 Args: experience: A batch of experience data in the ...
_loss
python
tensorflow/agents
tf_agents/agents/categorical_dqn/categorical_dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/categorical_dqn_agent.py
Apache-2.0
def _next_q_distribution(self, next_time_steps): """Compute the q distribution of the next state for TD error computation. Args: next_time_steps: A batch of next timesteps Returns: A [batch_size, num_atoms] tensor representing the Q-distribution for the next state. """ network_ob...
Compute the q distribution of the next state for TD error computation. Args: next_time_steps: A batch of next timesteps Returns: A [batch_size, num_atoms] tensor representing the Q-distribution for the next state.
_next_q_distribution
python
tensorflow/agents
tf_agents/agents/categorical_dqn/categorical_dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/categorical_dqn_agent.py
Apache-2.0
def project_distribution( supports: types.Tensor, weights: types.Tensor, target_support: types.Tensor, validate_args: bool = False, ) -> types.Tensor: """Projects a batch of (support, weights) onto target_support. Based on equation (7) in (Bellemare et al., 2017): https://arxiv.org/abs/1707.068...
Projects a batch of (support, weights) onto target_support. Based on equation (7) in (Bellemare et al., 2017): https://arxiv.org/abs/1707.06887 In the rest of the comments we will refer to this equation simply as Eq7. This code is not easy to digest, so we will use a running example to clarify what is goi...
project_distribution
python
tensorflow/agents
tf_agents/agents/categorical_dqn/categorical_dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/categorical_dqn_agent.py
Apache-2.0
def __init__( self, root_dir, env_name, num_iterations=200, max_episode_frames=108000, # ALE frames terminal_on_life_loss=False, conv_layer_params=((32, (8, 8), 4), (64, (4, 4), 2), (64, (3, 3), 1)), fc_layer_params=(512,), # Params for collect initial_collec...
A simple Atari train and eval for DQN. Args: root_dir: Directory to write log files to. env_name: Fully-qualified name of the Atari environment (i.e. Pong-v0). num_iterations: Number of train/eval iterations to run. max_episode_frames: Maximum length of a single episode, in ALE frames. ...
__init__
python
tensorflow/agents
tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
Apache-2.0
def _initial_collect(self): """Collect initial experience before training begins.""" logging.info('Collecting initial experience...') time_step_spec = ts.time_step_spec(self._env.observation_spec()) random_policy = random_py_policy.RandomPyPolicy( time_step_spec, self._env.action_spec() ) ...
Collect initial experience before training begins.
_initial_collect
python
tensorflow/agents
tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
Apache-2.0
def _collect_step(self, time_step, metric_observers, train=False): """Run a single step (or 2 steps on life loss) in the environment.""" if train: policy = self._collect_policy else: policy = self._eval_policy with self._action_timer: action_step = policy.action(time_step) with se...
Run a single step (or 2 steps on life loss) in the environment.
_collect_step
python
tensorflow/agents
tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
Apache-2.0
def _maybe_log(self, sess, global_step_val, total_loss): """Log some stats if global_step_val is a multiple of log_interval.""" if global_step_val % self._log_interval == 0: logging.info('step = %d, loss = %f', global_step_val, total_loss.loss) logging.info('%s', 'action_time = {}'.format(self._acti...
Log some stats if global_step_val is a multiple of log_interval.
_maybe_log
python
tensorflow/agents
tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
Apache-2.0
def get_run_args(): """Builds a dict of run arguments from flags.""" run_args = {} if FLAGS.num_iterations: run_args['num_iterations'] = FLAGS.num_iterations if FLAGS.initial_collect_steps: run_args['initial_collect_steps'] = FLAGS.initial_collect_steps if FLAGS.replay_buffer_capacity: run_args['r...
Builds a dict of run arguments from flags.
get_run_args
python
tensorflow/agents
tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/categorical_dqn/examples/train_eval_atari.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, critic_network: network.Network, actor_network: network.Network, actor_optimizer: types.Optimizer, critic_optimizer: types.Optimizer, alpha_optimizer: types.Optimizer, cql_alpha: U...
Creates a CQL-SAC Agent. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of BoundedTensorSpec representing the actions. critic_network: A function critic_network((observations, actions)) that returns the q_values for each observation and action. ...
__init__
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def _train(self, experience, weights): """Returns a train op to update the agent's networks. This method trains with the provided batched experience. Args: experience: A time-stacked trajectory object. weights: Optional scalar or elementwise (per-batch-entry) importance weights. R...
Returns a train op to update the agent's networks. This method trains with the provided batched experience. Args: experience: A time-stacked trajectory object. weights: Optional scalar or elementwise (per-batch-entry) importance weights. Returns: A train_op. Raises: V...
_train
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def _transpose_tile_and_batch_dims( self, original_tensor: types.Tensor ) -> types.Tensor: """Transposes [tile, batch, ...] to [batch, tile, ...].""" return tf.transpose( original_tensor, [1, 0] + list(range(2, len(original_tensor.shape))) )
Transposes [tile, batch, ...] to [batch, tile, ...].
_transpose_tile_and_batch_dims
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def _actions_and_log_probs( self, time_steps: ts.TimeStep, training: Optional[bool] = False ) -> Tuple[types.Tensor, types.Tensor]: """Get actions and corresponding log probabilities from policy.""" # Get raw action distribution from policy, and initialize bijectors list. batch_size = nest_utils.get...
Get actions and corresponding log probabilities from policy.
_actions_and_log_probs
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def _sample_and_transpose_actions_and_log_probs( self, time_steps: ts.TimeStep, num_action_samples: int, training: Optional[bool] = False, ) -> Tuple[types.Tensor, types.Tensor]: """Samples actions and corresponding log probabilities from policy.""" # Get raw action distribution from p...
Samples actions and corresponding log probabilities from policy.
_sample_and_transpose_actions_and_log_probs
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def _flattened_multibatch_tensor( self, original_tensor: types.Tensor ) -> types.Tensor: """Flattens the batch and tile dimensions into a single dimension. Args: original_tensor: Input tensor of shape [batch_size, tile, dim]. Returns: Flattened tensor with the outer dimension (batch_si...
Flattens the batch and tile dimensions into a single dimension. Args: original_tensor: Input tensor of shape [batch_size, tile, dim]. Returns: Flattened tensor with the outer dimension (batch_size * tile).
_flattened_multibatch_tensor
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def _get_q_values( self, target_input: Tuple[types.Tensor, types.Tensor], step_type: types.Tensor, reshape_batch_size: Optional[int], training: Optional[bool] = False, ) -> Tuple[types.Tensor, types.Tensor]: """Gets the Q-values of target_input. Uses the smaller of the critic ne...
Gets the Q-values of target_input. Uses the smaller of the critic network outputs since learned Q functions can overestimate Q-values. Args: target_input: Tuple of (observation, sampled actions) tensors. step_type: `Tensor` of `StepType` enum values. reshape_batch_size: Batch size to res...
_get_q_values
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def _cql_loss( self, time_steps: ts.TimeStep, actions: types.Tensor, training: Optional[bool] = False, ) -> types.Tensor: """Computes CQL loss for SAC training in continuous action spaces. Extends the standard critic loss to minimize Q-values sampled from a policy and maximize val...
Computes CQL loss for SAC training in continuous action spaces. Extends the standard critic loss to minimize Q-values sampled from a policy and maximize values of the dataset actions. Based on the `CQL(H)` equation (4) in (Kumar et al., 2020): ``` log_sum_exp(Q(s, a')) - Q(s, a) ``` Othe...
_cql_loss
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def actor_loss( self, time_steps: ts.TimeStep, actions: types.Tensor, weights: Optional[types.Tensor] = None, training: Optional[bool] = True, ) -> types.Tensor: """Computes actor_loss equivalent to the SAC actor_loss. Uses behavioral cloning for the first `self._num_bc_steps` o...
Computes actor_loss equivalent to the SAC actor_loss. Uses behavioral cloning for the first `self._num_bc_steps` of training. Args: time_steps: A batch of timesteps. actions: A batch of actions. weights: Optional scalar or elementwise (per-batch-entry) importance weights. train...
actor_loss
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def _critic_loss_with_optional_entropy_term( self, time_steps: ts.TimeStep, actions: types.Tensor, next_time_steps: ts.TimeStep, td_errors_loss_fn: types.LossFn, gamma: types.Float = 1.0, reward_scale_factor: types.Float = 1.0, weights: Optional[types.Tensor] = None, ...
Computes the critic loss for CQL-SAC training. The original SAC critic loss is: ``` (q(s, a) - (r(s, a) + \gamma q(s', a') - \gamma \alpha \log \pi(a'|s')))^2 ``` The CQL-SAC critic loss makes the entropy term optional. CQL may value unseen actions higher since it lower-bounds the value of ...
_critic_loss_with_optional_entropy_term
python
tensorflow/agents
tf_agents/agents/cql/cql_sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/cql/cql_sac_agent.py
Apache-2.0
def __init__( self, input_tensor_spec, output_tensor_spec, fc_layer_params=None, dropout_layer_params=None, conv_layer_params=None, activation_fn=tf.keras.activations.relu, kernel_initializer=None, last_kernel_initializer=None, name='ActorNetwork', ): ""...
Creates an instance of `ActorNetwork`. Args: input_tensor_spec: A nest of `tensor_spec.TensorSpec` representing the inputs. output_tensor_spec: A nest of `tensor_spec.BoundedTensorSpec` representing the outputs. fc_layer_params: Optional list of fully_connected parameters, where e...
__init__
python
tensorflow/agents
tf_agents/agents/ddpg/actor_network.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/actor_network.py
Apache-2.0
def __init__( self, input_tensor_spec, output_tensor_spec, conv_layer_params=None, input_fc_layer_params=(200, 100), lstm_size=(40,), output_fc_layer_params=(200, 100), activation_fn=tf.keras.activations.relu, name='ActorRnnNetwork', ): """Creates an instance ...
Creates an instance of `ActorRnnNetwork`. Args: input_tensor_spec: A nest of `tensor_spec.TensorSpec` representing the input observations. output_tensor_spec: A nest of `tensor_spec.BoundedTensorSpec` representing the actions. conv_layer_params: Optional list of convolution layers...
__init__
python
tensorflow/agents
tf_agents/agents/ddpg/actor_rnn_network.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/actor_rnn_network.py
Apache-2.0
def __init__( self, input_tensor_spec, observation_conv_layer_params=None, observation_fc_layer_params=None, observation_dropout_layer_params=None, action_fc_layer_params=None, action_dropout_layer_params=None, joint_fc_layer_params=None, joint_dropout_layer_params=...
Creates an instance of `CriticNetwork`. Args: input_tensor_spec: A tuple of (observation, action) each a nest of `tensor_spec.TensorSpec` representing the inputs. observation_conv_layer_params: Optional list of convolution layer parameters for observations, where each item is a length-t...
__init__
python
tensorflow/agents
tf_agents/agents/ddpg/critic_network.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/critic_network.py
Apache-2.0
def __init__( self, input_tensor_spec, observation_conv_layer_params=None, observation_fc_layer_params=(200,), action_fc_layer_params=(200,), joint_fc_layer_params=(100,), lstm_size=None, output_fc_layer_params=(200, 100), activation_fn=tf.keras.activations.relu, ...
Creates an instance of `CriticRnnNetwork`. Args: input_tensor_spec: A tuple of (observation, action) each of type `tensor_spec.TensorSpec` representing the inputs. observation_conv_layer_params: Optional list of convolution layers parameters to apply to the observations, where each item...
__init__
python
tensorflow/agents
tf_agents/agents/ddpg/critic_rnn_network.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/critic_rnn_network.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, actor_network: network.Network, critic_network: network.Network, actor_optimizer: Optional[types.Optimizer] = None, critic_optimizer: Optional[types.Optimizer] = None, ou_stddev: types.F...
Creates a DDPG Agent. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of BoundedTensorSpec representing the actions. actor_network: A tf_agents.network.Network to be used by the agent. The network will be called with call(observation, step_type[, po...
__init__
python
tensorflow/agents
tf_agents/agents/ddpg/ddpg_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/ddpg_agent.py
Apache-2.0
def _get_target_updater(self, tau=1.0, period=1): """Performs a soft update of the target network parameters. For each weight w_s in the original network, and its corresponding weight w_t in the target network, a soft update is: w_t = (1- tau) x w_t + tau x ws Args: tau: A float scalar in [0...
Performs a soft update of the target network parameters. For each weight w_s in the original network, and its corresponding weight w_t in the target network, a soft update is: w_t = (1- tau) x w_t + tau x ws Args: tau: A float scalar in [0, 1]. Default `tau=1.0` means hard update. period: ...
_get_target_updater
python
tensorflow/agents
tf_agents/agents/ddpg/ddpg_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/ddpg_agent.py
Apache-2.0
def critic_loss( self, time_steps: ts.TimeStep, actions: types.NestedTensor, next_time_steps: ts.TimeStep, weights: Optional[types.Tensor] = None, training: bool = False, ) -> types.Tensor: """Computes the critic loss for DDPG training. Args: time_steps: A batch of t...
Computes the critic loss for DDPG training. Args: time_steps: A batch of timesteps. actions: A batch of actions. next_time_steps: A batch of next timesteps. weights: Optional scalar or element-wise (per-batch-entry) importance weights. training: Whether this loss is being used...
critic_loss
python
tensorflow/agents
tf_agents/agents/ddpg/ddpg_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/ddpg_agent.py
Apache-2.0
def actor_loss( self, time_steps: ts.TimeStep, weights: Optional[types.Tensor] = None, training: bool = False, ) -> types.Tensor: """Computes the actor_loss for DDPG training. Args: time_steps: A batch of timesteps. weights: Optional scalar or element-wise (per-batch-entry...
Computes the actor_loss for DDPG training. Args: time_steps: A batch of timesteps. weights: Optional scalar or element-wise (per-batch-entry) importance weights. training: Whether this loss is being used for training. Returns: actor_loss: A scalar actor loss.
actor_loss
python
tensorflow/agents
tf_agents/agents/ddpg/ddpg_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/ddpg_agent.py
Apache-2.0
def train_eval( root_dir, env_name='HalfCheetah-v2', eval_env_name=None, env_load_fn=suite_mujoco.load, num_iterations=2000000, actor_fc_layers=(400, 300), critic_obs_fc_layers=(400,), critic_action_fc_layers=None, critic_joint_fc_layers=(300,), # Params for collect initial_c...
A simple train and eval for DDPG.
train_eval
python
tensorflow/agents
tf_agents/agents/ddpg/examples/v2/train_eval.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/examples/v2/train_eval.py
Apache-2.0
def create_actor_network(fc_layer_units, action_spec): """Create an actor network for DDPG.""" flat_action_spec = tf.nest.flatten(action_spec) if len(flat_action_spec) > 1: raise ValueError('Only a single action tensor is supported by this network') flat_action_spec = flat_action_spec[0] fc_layers = [den...
Create an actor network for DDPG.
create_actor_network
python
tensorflow/agents
tf_agents/agents/ddpg/examples/v2/train_eval.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/examples/v2/train_eval.py
Apache-2.0
def train_eval( root_dir, env_name='cartpole', task_name='balance', observations_allowlist='position', num_iterations=100000, actor_fc_layers=(400, 300), actor_output_fc_layers=(100,), actor_lstm_size=(40,), critic_obs_fc_layers=(400,), critic_action_fc_layers=None, critic_jo...
A simple train and eval for DDPG.
train_eval
python
tensorflow/agents
tf_agents/agents/ddpg/examples/v2/train_eval_rnn.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/examples/v2/train_eval_rnn.py
Apache-2.0
def _check_network_output(self, net, label): """Check outputs of q_net and target_q_net against expected shape. Subclasses that require different q_network outputs should override this function. Args: net: A `Network`. label: A label to print in case of a mismatch. """ outputs = ne...
Check outputs of q_net and target_q_net against expected shape. Subclasses that require different q_network outputs should override this function. Args: net: A `Network`. label: A label to print in case of a mismatch.
_check_network_output
python
tensorflow/agents
tf_agents/agents/dqn/dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/dqn/dqn_agent.py
Apache-2.0
def _get_target_updater(self, tau=1.0, period=1): """Performs a soft update of the target network parameters. For each weight w_s in the q network, and its corresponding weight w_t in the target_q_network, a soft update is: w_t = (1 - tau) * w_t + tau * w_s Args: tau: A float scalar in [0, 1...
Performs a soft update of the target network parameters. For each weight w_s in the q network, and its corresponding weight w_t in the target_q_network, a soft update is: w_t = (1 - tau) * w_t + tau * w_s Args: tau: A float scalar in [0, 1]. Default `tau=1.0` means hard update. period: Ste...
_get_target_updater
python
tensorflow/agents
tf_agents/agents/dqn/dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/dqn/dqn_agent.py
Apache-2.0
def _loss( self, experience, td_errors_loss_fn=None, gamma=1.0, reward_scale_factor=1.0, weights=None, training=False, ): """Computes loss for DQN training. Args: experience: A batch of experience data in the form of a `Trajectory` or `Transition`. The ...
Computes loss for DQN training. Args: experience: A batch of experience data in the form of a `Trajectory` or `Transition`. The structure of `experience` must match that of `self.collect_policy.step_spec`. If a `Trajectory`, all tensors in `experience` must be shaped `[B, T, ...]` wh...
_loss
python
tensorflow/agents
tf_agents/agents/dqn/dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/dqn/dqn_agent.py
Apache-2.0
def _compute_next_q_values(self, next_time_steps, info): """Compute the q value of the next state for TD error computation. Args: next_time_steps: A batch of next timesteps info: PolicyStep.info that may be used by other agents inherited from dqn_agent. Returns: A tensor of Q val...
Compute the q value of the next state for TD error computation. Args: next_time_steps: A batch of next timesteps info: PolicyStep.info that may be used by other agents inherited from dqn_agent. Returns: A tensor of Q values for the given next state.
_compute_next_q_values
python
tensorflow/agents
tf_agents/agents/dqn/dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/dqn/dqn_agent.py
Apache-2.0
def _compute_next_q_values(self, next_time_steps, info): """Compute the q value of the next state for TD error computation. Args: next_time_steps: A batch of next timesteps info: PolicyStep.info that may be used by other agents inherited from dqn_agent. Returns: A tensor of Q val...
Compute the q value of the next state for TD error computation. Args: next_time_steps: A batch of next timesteps info: PolicyStep.info that may be used by other agents inherited from dqn_agent. Returns: A tensor of Q values for the given next state.
_compute_next_q_values
python
tensorflow/agents
tf_agents/agents/dqn/dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/dqn/dqn_agent.py
Apache-2.0
def _compute_next_q_values(self, next_time_steps, info): """Compute the q value of the next state for TD error computation. Args: next_time_steps: A batch of next timesteps info: PolicyStep.info that may be used by other agents inherited from dqn_agent. Returns: A tensor of Q val...
Compute the q value of the next state for TD error computation. Args: next_time_steps: A batch of next timesteps info: PolicyStep.info that may be used by other agents inherited from dqn_agent. Returns: A tensor of Q values for the given next state.
_compute_next_q_values
python
tensorflow/agents
tf_agents/agents/dqn/dqn_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/dqn/dqn_agent.py
Apache-2.0
def testLossNStepMidMidLastFirst(self, agent_class): """Tests that n-step loss handles LAST time steps properly.""" q_net = DummyNet(self._observation_spec, self._action_spec) agent = agent_class( self._time_step_spec, self._action_spec, q_network=q_net, optimizer=None, ...
Tests that n-step loss handles LAST time steps properly.
testLossNStepMidMidLastFirst
python
tensorflow/agents
tf_agents/agents/dqn/dqn_agent_test.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/dqn/dqn_agent_test.py
Apache-2.0
def train_eval( root_dir, env_name='CartPole-v0', num_iterations=100000, train_sequence_length=1, # Params for QNetwork fc_layer_params=(100,), # Params for QRnnNetwork input_fc_layer_params=(50,), lstm_size=(20,), output_fc_layer_params=(20,), # Params for collect initia...
A simple train and eval for DQN.
train_eval
python
tensorflow/agents
tf_agents/agents/dqn/examples/v2/train_eval.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/dqn/examples/v2/train_eval.py
Apache-2.0
def tanh_and_scale_to_spec(inputs, spec): """Maps inputs with arbitrary range to range defined by spec using `tanh`.""" means = (spec.maximum + spec.minimum) / 2.0 magnitudes = (spec.maximum - spec.minimum) / 2.0 return means + magnitudes * tf.tanh(inputs)
Maps inputs with arbitrary range to range defined by spec using `tanh`.
tanh_and_scale_to_spec
python
tensorflow/agents
tf_agents/agents/ppo/ppo_actor_network.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_actor_network.py
Apache-2.0
def create_sequential_actor_net( self, fc_layer_units, action_tensor_spec, seed=None ): """Helper method for creating the actor network.""" self._seed_stream = self.seed_stream_class( seed=seed, salt='tf_agents_sequential_layers' ) def _get_seed(): seed = self._seed_stream() ...
Helper method for creating the actor network.
create_sequential_actor_net
python
tensorflow/agents
tf_agents/agents/ppo/ppo_actor_network.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_actor_network.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, optimizer: Optional[types.Optimizer] = None, actor_net: Optional[network.Network] = None, value_net: Optional[network.Network] = None, greedy_eval: bool = True, importance_ratio_clipping...
Creates a PPO Agent. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of `BoundedTensorSpec` representing the actions. optimizer: Optimizer to use for the agent, default to using `tf.compat.v1.train.AdamOptimizer`. actor_net: A `network.Distrib...
__init__
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def compute_advantages( self, rewards: types.NestedTensor, returns: types.Tensor, discounts: types.Tensor, value_preds: types.Tensor, ) -> types.Tensor: """Compute advantages, optionally using GAE. Based on baselines ppo1 implementation. Removes final timestep, as it needs t...
Compute advantages, optionally using GAE. Based on baselines ppo1 implementation. Removes final timestep, as it needs to use this timestep for next-step value prediction for TD error computation. Args: rewards: Tensor of per-timestep rewards. returns: Tensor of per-timestep returns. ...
compute_advantages
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def get_loss( self, time_steps: ts.TimeStep, actions: types.NestedTensorSpec, act_log_probs: types.Tensor, returns: types.Tensor, normalized_advantages: types.Tensor, action_distribution_parameters: types.NestedTensor, weights: types.Tensor, train_step: tf.Variable,...
Compute the loss and create optimization op for one training epoch. All tensors should have a single batch dimension. Args: time_steps: A minibatch of TimeStep tuples. actions: A minibatch of actions. act_log_probs: A minibatch of action probabilities (probability under the sampling ...
get_loss
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def compute_return_and_advantage( self, next_time_steps: ts.TimeStep, value_preds: types.Tensor ) -> Tuple[types.Tensor, types.Tensor]: """Compute the Monte Carlo return and advantage. Args: next_time_steps: batched tensor of TimeStep tuples after action is taken. value_preds: Batched value...
Compute the Monte Carlo return and advantage. Args: next_time_steps: batched tensor of TimeStep tuples after action is taken. value_preds: Batched value prediction tensor. Should have one more entry in time index than time_steps, with the final value corresponding to the value predictio...
compute_return_and_advantage
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def _preprocess(self, experience): """Performs advantage calculation for the collected experience. Args: experience: A (batch of) experience in the form of a `Trajectory`. The structure of `experience` must match that of `self.collect_data_spec`. All tensors in `experience` must be shaped...
Performs advantage calculation for the collected experience. Args: experience: A (batch of) experience in the form of a `Trajectory`. The structure of `experience` must match that of `self.collect_data_spec`. All tensors in `experience` must be shaped `[batch, time + 1, ...]` or [time...
_preprocess
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def _preprocess_sequence(self, experience): """Performs advantage calculation for the collected experience. This function is a no-op if self._compute_value_and_advantage_in_train is True, which means advantage calculation happens as part of agent.train(). Args: experience: A (batch of) experienc...
Performs advantage calculation for the collected experience. This function is a no-op if self._compute_value_and_advantage_in_train is True, which means advantage calculation happens as part of agent.train(). Args: experience: A (batch of) experience in the form of a `Trajectory`. The struct...
_preprocess_sequence
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def entropy_regularization_loss( self, time_steps: ts.TimeStep, entropy: types.Tensor, weights: types.Tensor, debug_summaries: bool = False, ) -> types.Tensor: """Create regularization loss tensor based on agent parameters.""" if self._entropy_regularization > 0: nest_utils...
Create regularization loss tensor based on agent parameters.
entropy_regularization_loss
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def value_estimation_loss( self, time_steps: ts.TimeStep, returns: types.Tensor, weights: types.Tensor, old_value_predictions: Optional[types.Tensor] = None, debug_summaries: bool = False, training: bool = False, ) -> types.Tensor: """Computes the value estimation loss fo...
Computes the value estimation loss for actor-critic training. All tensors should have a single batch dimension. Args: time_steps: A batch of timesteps. returns: Per-timestep returns for value function to predict. (Should come from TD-lambda computation.) weights: Optional scalar or e...
value_estimation_loss
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def policy_gradient_loss( self, time_steps: ts.TimeStep, actions: types.NestedTensor, sample_action_log_probs: types.Tensor, advantages: types.Tensor, current_policy_distribution: types.NestedDistribution, weights: types.Tensor, debug_summaries: bool = False, ) -> types...
Create tensor for policy gradient loss. All tensors should have a single batch dimension. Args: time_steps: TimeSteps with observations for each timestep. actions: Tensor of actions for timesteps, aligned on index. sample_action_log_probs: Tensor of sample probability of each action. a...
policy_gradient_loss
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def kl_penalty_loss( self, time_steps: ts.TimeStep, action_distribution_parameters: types.NestedTensor, current_policy_distribution: types.NestedDistribution, weights: types.Tensor, debug_summaries: bool = False, ) -> types.Tensor: """Compute a loss that penalizes policy steps ...
Compute a loss that penalizes policy steps with high KL. Based on KL divergence from old (data-collection) policy to new (updated) policy. All tensors should have a single batch dimension. Args: time_steps: TimeStep tuples with observations for each timestep. Used for computing new acti...
kl_penalty_loss
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def update_adaptive_kl_beta( self, kl_divergence: types.Tensor ) -> Optional[tf.Operation]: """Create update op for adaptive KL penalty coefficient. Args: kl_divergence: KL divergence of old policy to new policy for all timesteps. Returns: update_op: An op which runs the update...
Create update op for adaptive KL penalty coefficient. Args: kl_divergence: KL divergence of old policy to new policy for all timesteps. Returns: update_op: An op which runs the update for the adaptive kl penalty term.
update_adaptive_kl_beta
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def _get_discount(experience) -> types.Tensor: """Try to get the discount entry from `experience`. Typically experience is either a Trajectory or a Transition. Args: experience: Data collected from e.g. a replay buffer. Returns: discount: The discount tensor stored in `experience`. """ if isinsta...
Try to get the discount entry from `experience`. Typically experience is either a Trajectory or a Transition. Args: experience: Data collected from e.g. a replay buffer. Returns: discount: The discount tensor stored in `experience`.
_get_discount
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent.py
Apache-2.0
def _compute_returns_fn(rewards, discounts, next_state_return=0.0): """Python implementation of computing discounted returns.""" returns = np.zeros_like(rewards) for t in range(len(returns) - 1, -1, -1): returns[t] = rewards[t] + discounts[t] * next_state_return next_state_return = returns[t] return ret...
Python implementation of computing discounted returns.
_compute_returns_fn
python
tensorflow/agents
tf_agents/agents/ppo/ppo_agent_test.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_agent_test.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, optimizer: Optional[types.Optimizer] = None, actor_net: Optional[network.Network] = None, value_net: Optional[network.Network] = None, greedy_eval: bool = True, importance_ratio_clipping...
Creates a PPO Agent implementing the clipped probability ratios. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of BoundedTensorSpec representing the actions. optimizer: Optimizer to use for the agent. actor_net: A function actor_net(observations, ac...
__init__
python
tensorflow/agents
tf_agents/agents/ppo/ppo_clip_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_clip_agent.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, actor_net: network.Network, value_net: network.Network, num_epochs: int, initial_adaptive_kl_beta: types.Float, adaptive_kl_target: types.Float, adaptive_kl_tolerance: types.Float,...
Creates a PPO Agent implementing the KL penalty loss. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of `BoundedTensorSpec` representing the actions. actor_net: A `network.DistributionNetwork` which maps observations to action distributions. Common...
__init__
python
tensorflow/agents
tf_agents/agents/ppo/ppo_kl_penalty_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_kl_penalty_agent.py
Apache-2.0
def get_initial_value_state( self, batch_size: types.Int ) -> types.NestedTensor: """Returns the initial state of the value network. Args: batch_size: A constant or Tensor holding the batch size. Can be None, in which case the state will not have a batch dimension added. Returns: ...
Returns the initial state of the value network. Args: batch_size: A constant or Tensor holding the batch size. Can be None, in which case the state will not have a batch dimension added. Returns: A nest of zero tensors matching the spec of the value network state.
get_initial_value_state
python
tensorflow/agents
tf_agents/agents/ppo/ppo_policy.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_policy.py
Apache-2.0
def apply_value_network( self, observations: types.NestedTensor, step_types: types.Tensor, value_state: Optional[types.NestedTensor] = None, training: bool = False, ) -> types.NestedTensor: """Apply value network to time_step, potentially a sequence. If observation_normalizer is...
Apply value network to time_step, potentially a sequence. If observation_normalizer is not None, applies observation normalization. Args: observations: A (possibly nested) observation tensor with outer_dims either (batch_size,) or (batch_size, time_index). If observations is a time serie...
apply_value_network
python
tensorflow/agents
tf_agents/agents/ppo/ppo_policy.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_policy.py
Apache-2.0
def make_trajectory_mask(batched_traj: trajectory.Trajectory) -> types.Tensor: """Mask boundary trajectories and those with invalid returns and advantages. Args: batched_traj: Trajectory, doubly-batched [batch_dim, time_dim,...]. It must be preprocessed already. Returns: A mask, type tf.float32, t...
Mask boundary trajectories and those with invalid returns and advantages. Args: batched_traj: Trajectory, doubly-batched [batch_dim, time_dim,...]. It must be preprocessed already. Returns: A mask, type tf.float32, that is 0.0 for all between-episode Trajectory (batched_traj.step_type is LAST)...
make_trajectory_mask
python
tensorflow/agents
tf_agents/agents/ppo/ppo_utils.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_utils.py
Apache-2.0
def make_timestep_mask( batched_next_time_step: ts.TimeStep, allow_partial_episodes: bool = False ) -> types.Tensor: """Create a mask for transitions and optionally final incomplete episodes. Args: batched_next_time_step: Next timestep, doubly-batched [batch_dim, time_dim, ...]. allow_partial_epi...
Create a mask for transitions and optionally final incomplete episodes. Args: batched_next_time_step: Next timestep, doubly-batched [batch_dim, time_dim, ...]. allow_partial_episodes: If true, then steps on incomplete episodes are allowed. Returns: A mask, type tf.float32, that is 0.0 for ...
make_timestep_mask
python
tensorflow/agents
tf_agents/agents/ppo/ppo_utils.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_utils.py
Apache-2.0
def nested_kl_divergence( nested_from_distribution: types.NestedDistribution, nested_to_distribution: types.NestedDistribution, outer_dims: Sequence[int] = (), ) -> types.Tensor: """Given two nested distributions, sum the KL divergences of the leaves.""" nest_utils.assert_same_structure( nested_fr...
Given two nested distributions, sum the KL divergences of the leaves.
nested_kl_divergence
python
tensorflow/agents
tf_agents/agents/ppo/ppo_utils.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_utils.py
Apache-2.0
def get_metric_observers(metrics): """Returns a list of observers, one for each metric.""" def get_metric_observer(metric): def metric_observer(time_step, action, next_time_step, policy_state): action_step = policy_step.PolicyStep(action, policy_state, ()) traj = trajectory.from_transition(time_st...
Returns a list of observers, one for each metric.
get_metric_observers
python
tensorflow/agents
tf_agents/agents/ppo/ppo_utils.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_utils.py
Apache-2.0
def get_learning_rate(optimizer): """Gets the current learning rate from an optimizer to be graphed.""" # Keras optimizers uses `learning_rate`. if hasattr(optimizer, 'learning_rate'): learning_rate = optimizer.learning_rate # pylint: disable=protected-access # Adam optimizers store their learning rate in ...
Gets the current learning rate from an optimizer to be graphed.
get_learning_rate
python
tensorflow/agents
tf_agents/agents/ppo/ppo_utils.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/ppo_utils.py
Apache-2.0
def train_eval( root_dir, env_name='HalfCheetah-v2', env_load_fn=suite_mujoco.load, random_seed=None, # TODO(b/127576522): rename to policy_fc_layers. actor_fc_layers=(200, 100), value_fc_layers=(200, 100), use_rnns=False, lstm_size=(20,), # Params for collect num_environment...
A simple train and eval for PPO.
train_eval
python
tensorflow/agents
tf_agents/agents/ppo/examples/v2/train_eval_clip_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ppo/examples/v2/train_eval_clip_agent.py
Apache-2.0
def _get_target_updater(self, tau=1.0, period=1): """Performs a soft update of the target network. For each weight w_s in the q network, and its corresponding weight w_t in the target_q_network, a soft update is: w_t = (1 - tau) * w_t + tau * w_s Args: tau: A float scalar in [0, 1]. Default ...
Performs a soft update of the target network. For each weight w_s in the q network, and its corresponding weight w_t in the target_q_network, a soft update is: w_t = (1 - tau) * w_t + tau * w_s Args: tau: A float scalar in [0, 1]. Default `tau=1.0` means hard update. Used for target netw...
_get_target_updater
python
tensorflow/agents
tf_agents/agents/qtopt/qtopt_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/qtopt/qtopt_agent.py
Apache-2.0
def _get_target_updater_delayed(self, tau_delayed=1.0, period_delayed=1): """Performs a soft update of the delayed target network. For each weight w_s in the q network, and its corresponding weight w_t in the target_q_network, a soft update is: w_t = (1 - tau) * w_t + tau * w_s Args: tau_del...
Performs a soft update of the delayed target network. For each weight w_s in the q network, and its corresponding weight w_t in the target_q_network, a soft update is: w_t = (1 - tau) * w_t + tau * w_s Args: tau_delayed: A float scalar in [0, 1]. Default `tau=1.0` means hard update. Used...
_get_target_updater_delayed
python
tensorflow/agents
tf_agents/agents/qtopt/qtopt_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/qtopt/qtopt_agent.py
Apache-2.0
def _get_target_updater_delayed_2(self, tau_delayed=1.0, period_delayed=1): """Performs a soft update of the delayed target network. For each weight w_s in the q network, and its corresponding weight w_t in the target_q_network, a soft update is: w_t = (1 - tau) * w_t + tau * w_s Args: tau_d...
Performs a soft update of the delayed target network. For each weight w_s in the q network, and its corresponding weight w_t in the target_q_network, a soft update is: w_t = (1 - tau) * w_t + tau * w_s Args: tau_delayed: A float scalar in [0, 1]. Default `tau=1.0` means hard update. Used...
_get_target_updater_delayed_2
python
tensorflow/agents
tf_agents/agents/qtopt/qtopt_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/qtopt/qtopt_agent.py
Apache-2.0
def _add_auxiliary_losses(self, transition, weights, losses_dict): """Computes auxiliary losses, updating losses_dict in place.""" total_auxiliary_loss = 0 if self._auxiliary_loss_fns is not None: for auxiliary_loss_fn in self._auxiliary_loss_fns: auxiliary_loss, auxiliary_reg_loss = auxiliary...
Computes auxiliary losses, updating losses_dict in place.
_add_auxiliary_losses
python
tensorflow/agents
tf_agents/agents/qtopt/qtopt_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/qtopt/qtopt_agent.py
Apache-2.0
def _loss(self, experience, weights=None, training=False): """Computes loss for QtOpt training. Args: experience: A batch of experience data in the form of a `Trajectory` or `Transition`. The structure of `experience` must match that of `self.collect_policy.step_spec`. If a `Trajectory`,...
Computes loss for QtOpt training. Args: experience: A batch of experience data in the form of a `Trajectory` or `Transition`. The structure of `experience` must match that of `self.collect_policy.step_spec`. If a `Trajectory`, all tensors in `experience` must be shaped `[B, T, ...]` ...
_loss
python
tensorflow/agents
tf_agents/agents/qtopt/qtopt_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/qtopt/qtopt_agent.py
Apache-2.0
def VerifyTrainAndRestore(self): """Helper function for testing correct variable updating and restoring.""" batch_size = 2 seq_len = 2 observations = tensor_spec.sample_spec_nest( self._observation_spec, outer_dims=(batch_size, seq_len) ) actions = tensor_spec.sample_spec_nest( s...
Helper function for testing correct variable updating and restoring.
VerifyTrainAndRestore
python
tensorflow/agents
tf_agents/agents/qtopt/qtopt_agent_test.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/qtopt/qtopt_agent_test.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, policy_class: PolicyClassType, debug_summaries: bool = False, summarize_grads_and_vars: bool = False, train_step_counter: Optional[tf.Variable] = None, num_outer_dims: int = 1, nam...
Creates a fixed-policy agent with no-op for training. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of BoundedTensorSpec representing the actions. policy_class: a tf_policy.TFPolicy or py_policy.PyPolicy class to use as a policy. debug_summa...
__init__
python
tensorflow/agents
tf_agents/agents/random/fixed_policy_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/random/fixed_policy_agent.py
Apache-2.0
def _train(self, experience, weights): """Do nothing. Arguments are ignored and loss is always 0.""" del experience # Unused del weights # Unused # Incrementing the step counter. self.train_step_counter.assign_add(1) # Returning 0 loss. return tf_agent.LossInfo(0.0, None)
Do nothing. Arguments are ignored and loss is always 0.
_train
python
tensorflow/agents
tf_agents/agents/random/fixed_policy_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/random/fixed_policy_agent.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, debug_summaries: bool = False, summarize_grads_and_vars: bool = False, train_step_counter: Optional[tf.Variable] = None, num_outer_dims: int = 1, name: Optional[Text] = None, ): ""...
Creates a random agent. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of BoundedTensorSpec representing the actions. debug_summaries: A bool to gather debug summaries. summarize_grads_and_vars: If true, gradient summaries will be written. trai...
__init__
python
tensorflow/agents
tf_agents/agents/random/random_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/random/random_agent.py
Apache-2.0
def _standard_normalize(values, axes=(0,)): """Standard normalizes values `values`. Args: values: Tensor with values to be standardized. axes: Axes used to compute mean and variances. Returns: Standardized values (values - mean(values[axes])) / std(values[axes]). """ values_mean, values_var = tf...
Standard normalizes values `values`. Args: values: Tensor with values to be standardized. axes: Axes used to compute mean and variances. Returns: Standardized values (values - mean(values[axes])) / std(values[axes]).
_standard_normalize
python
tensorflow/agents
tf_agents/agents/reinforce/reinforce_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/reinforce/reinforce_agent.py
Apache-2.0
def _entropy_loss(distributions, spec, weights=None): """Computes entropy loss. Args: distributions: A possibly batched tuple of distributions. spec: A nested tuple representing the action spec. weights: Optional scalar or element-wise (per-batch-entry) importance weights. Includes a mask for in...
Computes entropy loss. Args: distributions: A possibly batched tuple of distributions. spec: A nested tuple representing the action spec. weights: Optional scalar or element-wise (per-batch-entry) importance weights. Includes a mask for invalid timesteps. Returns: A Tensor representing the ...
_entropy_loss
python
tensorflow/agents
tf_agents/agents/reinforce/reinforce_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/reinforce/reinforce_agent.py
Apache-2.0
def _get_initial_policy_state(policy, time_steps): """Gets the initial state of a policy.""" batch_size = ( tf.compat.dimension_at_index(time_steps.discount.shape, 0) or tf.shape(time_steps.discount)[0] ) return policy.get_initial_state(batch_size=batch_size)
Gets the initial state of a policy.
_get_initial_policy_state
python
tensorflow/agents
tf_agents/agents/reinforce/reinforce_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/reinforce/reinforce_agent.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.TensorSpec, actor_network: network.Network, optimizer: types.Optimizer, value_network: Optional[network.Network] = None, value_estimation_loss_coef: types.Float = 0.2, advantage_fn: Optional[AdvantageFnTy...
Creates a REINFORCE Agent. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of BoundedTensorSpec representing the actions. actor_network: A tf_agents.network.Network to be used by the agent. The network will be called with call(observation, step_type...
__init__
python
tensorflow/agents
tf_agents/agents/reinforce/reinforce_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/reinforce/reinforce_agent.py
Apache-2.0
def policy_gradient_loss( self, actions_distribution: types.NestedDistribution, actions: types.NestedTensor, is_boundary: types.Tensor, returns: types.Tensor, num_episodes: types.Int, weights: Optional[types.Tensor] = None, ) -> types.Tensor: """Computes the policy gradie...
Computes the policy gradient loss. Args: actions_distribution: A possibly batched tuple of action distributions. actions: Tensor with a batch of actions. is_boundary: Tensor of booleans that indicate if the corresponding action was in a boundary trajectory and should be ignored. ret...
policy_gradient_loss
python
tensorflow/agents
tf_agents/agents/reinforce/reinforce_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/reinforce/reinforce_agent.py
Apache-2.0
def entropy_regularization_loss( self, actions_distribution: types.NestedDistribution, weights: Optional[types.Tensor] = None, ) -> types.Tensor: """Computes the optional entropy regularization loss. Extending REINFORCE by entropy regularization was originally proposed in "Function opti...
Computes the optional entropy regularization loss. Extending REINFORCE by entropy regularization was originally proposed in "Function optimization using connectionist reinforcement learning algorithms." (Williams and Peng, 1991). Args: actions_distribution: A possibly batched tuple of action dis...
entropy_regularization_loss
python
tensorflow/agents
tf_agents/agents/reinforce/reinforce_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/reinforce/reinforce_agent.py
Apache-2.0
def value_estimation_loss( self, value_preds: types.Tensor, returns: types.Tensor, num_episodes: types.Int, weights: Optional[types.Tensor] = None, ) -> types.Tensor: """Computes the value estimation loss. Args: value_preds: Per-timestep estimated values. returns: Pe...
Computes the value estimation loss. Args: value_preds: Per-timestep estimated values. returns: Per-timestep returns for value function to predict. num_episodes: Number of episodes contained in the training data. weights: Optional scalar or element-wise (per-batch-entry) importance w...
value_estimation_loss
python
tensorflow/agents
tf_agents/agents/reinforce/reinforce_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/reinforce/reinforce_agent.py
Apache-2.0
def train_eval( root_dir, env_name='CartPole-v0', num_iterations=1000, actor_fc_layers=(100,), value_net_fc_layers=(100,), use_value_network=False, use_tf_functions=True, # Params for collect collect_episodes_per_iteration=2, replay_buffer_capacity=2000, # Params for train ...
A simple train and eval for Reinforce.
train_eval
python
tensorflow/agents
tf_agents/agents/reinforce/examples/v2/train_eval.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/reinforce/examples/v2/train_eval.py
Apache-2.0
def __init__( self, time_step_spec: ts.TimeStep, action_spec: types.NestedTensorSpec, critic_network: network.Network, actor_network: network.Network, actor_optimizer: types.Optimizer, critic_optimizer: types.Optimizer, alpha_optimizer: types.Optimizer, actor_loss_w...
Creates a SAC Agent. Args: time_step_spec: A `TimeStep` spec of the expected time_steps. action_spec: A nest of BoundedTensorSpec representing the actions. critic_network: A function critic_network((observations, actions)) that returns the q_values for each observation and action. a...
__init__
python
tensorflow/agents
tf_agents/agents/sac/sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/sac/sac_agent.py
Apache-2.0
def _initialize(self): """Returns an op to initialize the agent. Copies weights from the Q networks to the target Q network. """ common.soft_variables_update( self._critic_network_1.variables, self._target_critic_network_1.variables, tau=1.0, ) common.soft_variables_upda...
Returns an op to initialize the agent. Copies weights from the Q networks to the target Q network.
_initialize
python
tensorflow/agents
tf_agents/agents/sac/sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/sac/sac_agent.py
Apache-2.0
def _train(self, experience, weights): """Returns a train op to update the agent's networks. This method trains with the provided batched experience. Args: experience: A time-stacked trajectory object. weights: Optional scalar or elementwise (per-batch-entry) importance weights. R...
Returns a train op to update the agent's networks. This method trains with the provided batched experience. Args: experience: A time-stacked trajectory object. weights: Optional scalar or elementwise (per-batch-entry) importance weights. Returns: A train_op. Raises: V...
_train
python
tensorflow/agents
tf_agents/agents/sac/sac_agent.py
https://github.com/tensorflow/agents/blob/master/tf_agents/agents/sac/sac_agent.py
Apache-2.0