Small fix in epistemic neural networks

fbcode//pearl

Announcing Pearl serialization

Test serialization

dqn for frozen lake tutorial

Make all Pearl components torch.nn.Modules

Implement compare for safety modules

Implement compare for contextual bandits

Implement compare for action representation modules

Implement compare for (neural) linear regression modules

Refactor TestAgentWithPyTorch so its models can be reused

Unit test for dqn and double dqn

Publishing the model with exploration scores for CB use-cases

add the missing .shape to q value networks

Add missing Pyre mode headers] [batch:21/1502] [shard:31/N]

Documentation generation via gh actions (#107)

Allow twin critic to take network instances and test SAC in atari games.

Add CNNActorNetwork

Rename VanillaCNN to CNNValueNetwork and divide state values by 255.

add autotune to discrete sac

multi-head CNN

modify epsilon greedy exploration to add epsilon scheduling

Fix training_epochs meaning in Pearl offline learning

Multiple minor warnings

Fix the error when max_number_of_actions is None and the action_space…

setup atari experiments

allow passing optimizers to policy learners

change observations only from float64 to float32 before adding to the…

replace _critic_update_count with _training_steps in td3

Make csac's autotune more efficient