in tf_agents/bandits/policies/neural_linucb_policy.py [0:0]
def __init__(self,
encoding_network: types.Network,
encoding_dim: int,
reward_layer: tf.keras.layers.Dense,
epsilon_greedy: float,
actions_from_reward_layer: types.Bool,
cov_matrix: Sequence[types.Float],
data_vector: Sequence[types.Float],
num_samples: Sequence[types.Int],
time_step_spec: types.TimeStep,
alpha: float = 1.0,
emit_policy_info: Sequence[Text] = (),
emit_log_probability: bool = False,
accepts_per_arm_features: bool = False,
distributed_use_reward_layer: bool = False,
observation_and_action_constraint_splitter: Optional[
types.Splitter] = None,
name: Optional[Text] = None):
"""Initializes `NeuralLinUCBPolicy`.
Args:
encoding_network: network that encodes the observations.
encoding_dim: (int) dimension of the encoded observations.
reward_layer: final layer that predicts the expected reward per arm. In
case the policy accepts per-arm features, the output of this layer has
to be a scalar. This is because in the per-arm case, all encoded
observations have to go through the same computation to get the reward
estimates. The `num_actions` dimension of the encoded observation is
treated as a batch dimension in the reward layer.
epsilon_greedy: (float) representing the probability of choosing a random
action instead of the greedy action.
actions_from_reward_layer: (boolean variable) whether to get actions from
the reward layer or from LinUCB.
cov_matrix: list of the covariance matrices. There exists one covariance
matrix per arm, unless the policy accepts per-arm features, in which
case this list must have a single element.
data_vector: list of the data vectors. A data vector is a weighted sum
of the observations, where the weight is the corresponding reward. Each
arm has its own data vector, unless the policy accepts per-arm features,
in which case this list must have a single element.
num_samples: list of number of samples per arm. If the policy accepts per-
arm features, this is a single-element list counting the number of
steps.
time_step_spec: A `TimeStep` spec of the expected time_steps.
alpha: (float) non-negative weight multiplying the confidence intervals.
emit_policy_info: (tuple of strings) what side information we want to get
as part of the policy info. Allowed values can be found in
`policy_utilities.PolicyInfo`.
emit_log_probability: (bool) whether to emit log probabilities.
accepts_per_arm_features: (bool) Whether the policy accepts per-arm
features.
distributed_use_reward_layer: (bool) Whether to pick the actions using
the network or use LinUCB. This applies only in distributed training
setting and has a similar role to the `actions_from_reward_layer`
mentioned above.
observation_and_action_constraint_splitter: A function used for masking
valid/invalid actions with each state of the environment. The function
takes in a full observation and returns a tuple consisting of 1) the
part of the observation intended as input to the bandit policy and 2)
the mask. The mask should be a 0-1 `Tensor` of shape
`[batch_size, num_actions]`. This function should also work with a
`TensorSpec` as input, and should output `TensorSpec` objects for the
observation and mask.
name: The name of this policy.
"""
policy_utilities.check_no_mask_with_arm_features(
accepts_per_arm_features, observation_and_action_constraint_splitter)
encoding_network.create_variables()
self._encoding_network = encoding_network
self._reward_layer = reward_layer
self._encoding_dim = encoding_dim
if accepts_per_arm_features and reward_layer.units != 1:
raise ValueError('The output dimension of the reward layer must be 1, got'
' {}'.format(reward_layer.units))
if not isinstance(cov_matrix, (list, tuple)):
raise ValueError('cov_matrix must be a list of matrices (Tensors).')
self._cov_matrix = cov_matrix
if not isinstance(data_vector, (list, tuple)):
raise ValueError('data_vector must be a list of vectors (Tensors).')
self._data_vector = data_vector
if not isinstance(num_samples, (list, tuple)):
raise ValueError('num_samples must be a list of vectors (Tensors).')
self._num_samples = num_samples
self._alpha = alpha
self._actions_from_reward_layer = actions_from_reward_layer
self._epsilon_greedy = epsilon_greedy
self._dtype = self._data_vector[0].dtype
self._distributed_use_reward_layer = distributed_use_reward_layer
if len(cov_matrix) != len(data_vector):
raise ValueError('The size of list cov_matrix must match the size of '
'list data_vector. Got {} for cov_matrix and {} '
'for data_vector'.format(
len(self._cov_matrix), len((data_vector))))
if len(num_samples) != len(cov_matrix):
raise ValueError('The size of num_samples must match the size of '
'list cov_matrix. Got {} for num_samples and {} '
'for cov_matrix'.format(
len(self._num_samples), len((cov_matrix))))
self._accepts_per_arm_features = accepts_per_arm_features
if observation_and_action_constraint_splitter is not None:
context_spec, _ = observation_and_action_constraint_splitter(
time_step_spec.observation)
else:
context_spec = time_step_spec.observation
if accepts_per_arm_features:
self._num_actions = tf.nest.flatten(context_spec[
bandit_spec_utils.PER_ARM_FEATURE_KEY])[0].shape.as_list()[0]
self._num_models = 1
else:
self._num_actions = len(cov_matrix)
self._num_models = self._num_actions
cov_matrix_dim = tf.compat.dimension_value(cov_matrix[0].shape[0])
if self._encoding_dim != cov_matrix_dim:
raise ValueError('The dimension of matrix `cov_matrix` must match '
'encoding dimension {}.'
'Got {} for `cov_matrix`.'.format(
self._encoding_dim, cov_matrix_dim))
data_vector_dim = tf.compat.dimension_value(data_vector[0].shape[0])
if self._encoding_dim != data_vector_dim:
raise ValueError('The dimension of vector `data_vector` must match '
'encoding dimension {}. '
'Got {} for `data_vector`.'.format(
self._encoding_dim, data_vector_dim))
action_spec = tensor_spec.BoundedTensorSpec(
shape=(),
dtype=tf.int32,
minimum=0,
maximum=self._num_actions - 1,
name='action')
self._emit_policy_info = emit_policy_info
predicted_rewards_mean = ()
if policy_utilities.InfoFields.PREDICTED_REWARDS_MEAN in emit_policy_info:
predicted_rewards_mean = tensor_spec.TensorSpec(
[self._num_actions],
dtype=tf.float32)
predicted_rewards_optimistic = ()
if (policy_utilities.InfoFields.PREDICTED_REWARDS_OPTIMISTIC in
emit_policy_info):
predicted_rewards_optimistic = tensor_spec.TensorSpec(
[self._num_actions],
dtype=tf.float32)
if accepts_per_arm_features:
chosen_arm_features_info_spec = (
policy_utilities.create_chosen_arm_features_info_spec(
time_step_spec.observation))
info_spec = policy_utilities.PerArmPolicyInfo(
predicted_rewards_mean=predicted_rewards_mean,
predicted_rewards_optimistic=predicted_rewards_optimistic,
chosen_arm_features=chosen_arm_features_info_spec)
else:
info_spec = policy_utilities.PolicyInfo(
predicted_rewards_mean=predicted_rewards_mean,
predicted_rewards_optimistic=predicted_rewards_optimistic)
super(NeuralLinUCBPolicy, self).__init__(
time_step_spec=time_step_spec,
action_spec=action_spec,
emit_log_probability=emit_log_probability,
observation_and_action_constraint_splitter=(
observation_and_action_constraint_splitter),
info_spec=info_spec,
name=name)