tf_agents.agents.ppo.ppo_utils.distribution_from_spec

Convert (spec, new_distribution_params) -> Distribution.

tf_agents.agents.ppo.ppo_utils.distribution_from_spec(
    spec: tf_agents.typing.types.NestedTensorSpec,
    new_distribution_params: tf_agents.typing.types.NestedTensor,
    legacy_distribution_network: bool
) -> tf_agents.typing.types.NestedDistribution

new_distribution_params typically comes from a logged policy info.

spec A nested tensor spec. If legacy_distribution_network is True, these are typically actor_net.output_spec. If it's False, these are typically the output of actor_net.create_variables().
new_distribution_params Parameters to merge with the spec to create a new distribution. These were typically emitted by get_distribution_params and stored in the replay buffer.
legacy_distribution_network True if the spec and params were generated from a network.DistributionNetwork.

A (possibly nested set of) Distribution created from the spec merged with the new params.