rainbow_network.md

atari_lib.rainbow_network

atari_lib.rainbow_network(
    num_actions,
    num_atoms,
    support,
    network_type,
    state
)

The convolutional network used to compute agent's Q-value distributions.

Args:

• num_actions: int, number of actions.
• num_atoms: int, the number of buckets of the value function distribution.
• support: tf.linspace, the support of the Q-value distribution.
• network_type: namedtuple, collection of expected values to return.
• state: tf.Tensor, contains the agent's current state.

Returns:

• net: _network_type object containing the tensors output by the network.