Scalarize a rank-3 objectives tensor into a rank-2 tensor.
@tf.function
tf_agents.bandits.policies.greedy_multi_objective_neural_policy.scalarize_objectives(
objectives_tensor: tf.Tensor,
scalarizer: tf_agents.bandits.multi_objective.multi_objective_scalarizer.Scalarizer
)
Scalarize an objective values tensor shaped as
[batch_size, num_of_objectives, num_of_actions] along the second dimension
into a rank-2 tensor shaped as [batch_size, num_of_actions]
Returns |
A rank-2 tensor of scalarized rewards shaped as
[batch_size, num_of_actions].
|
Raises |
ValueError
|
If objectives_tensor is not rank-3.
|