|View source on GitHub|
Networks that take as input global and per-arm features, and output rewards.
class GlobalAndArmCommonTowerNetwork: A network that takes global and arm observations and outputs rewards.
class GlobalAndArmDotProductNetwork: A network that takes global and arm observations and outputs rewards.
create_feed_forward_common_tower_network(...): Creates a common tower network with feedforward towers.
create_feed_forward_dot_product_network(...): Creates a dot product network with feedforward towers.