academia.utils.models module

Ready-made network architectures to use with agents and environments provided by academia package.

See also

Note

These are just example architectures that allow users to quickly set up simple experiments. Users are encouraged to design their own network architectures that better suit their needs.

Models for BridgeBuilding environment

See also

academia.environments.BridgeBuilding

class academia.utils.models.bridge_building.MLPActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Bridge Building environment with the following parameters:

n_frames_stacked=1
append_step_count=False
obs_type='array'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.bridge_building.MLPCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Bridge Building environment with the following parameters:

n_frames_stacked=1
append_step_count=False
obs_type='array'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.bridge_building.MLPDQN

Bases: Module

Neural network architecture for DQN compatible with Bridge Building environment with the following parameters:

n_frames_stacked=1
append_step_count=False
obs_type='array'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.bridge_building.MLPStepActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Bridge Building environment with the following parameters:

n_frames_stacked=1
append_step_count=True
obs_type='array'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.bridge_building.MLPStepCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Bridge Building environment with the following parameters:

n_frames_stacked=1
append_step_count=True
obs_type='array'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.bridge_building.MLPStepDQN

Bases: Module

Neural network architecture for DQN compatible with Bridge Building environment with the following parameters:

n_frames_stacked=1
append_step_count=True
obs_type='array'

forward(obs) → Tensor: Defines the computation performed at every call.

Models for DoorKey environment

See also

academia.environments.DoorKey

class academia.utils.models.door_key.MLPActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Door Key environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.door_key.MLPCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Door Key environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.door_key.MLPDQN

Bases: Module

Neural network architecture for DQN compatible with Door Key environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(x) → Tensor: Defines the computation performed at every call.

class academia.utils.models.door_key.MLPStepActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Door Key environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.door_key.MLPStepCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Door Key environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.door_key.MLPStepDQN

Bases: Module

Neural network architecture for DQN compatible with Door Key environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(x) → Tensor: Defines the computation performed at every call.

Models for LavaCrossing environment

See also

academia.environments.LavaCrossing

class academia.utils.models.lava_crossing.MLPActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Lava Crossing environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lava_crossing.MLPCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Lava Crossing environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lava_crossing.MLPDQN

Bases: Module

Neural network architecture for DQN compatible with Lava Crossing environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(x) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lava_crossing.MLPStepActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Lava Crossing environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lava_crossing.MLPStepCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Lava Crossing environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lava_crossing.MLPStepDQN

Bases: Module

Neural network architecture for DQN compatible with Lava Crossing environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(x) → Tensor: Defines the computation performed at every call.

Models for LunarLander environment

See also

academia.environments.LunarLander

class academia.utils.models.lunar_lander.MLPActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Lunar Lander environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lunar_lander.MLPCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Lunar Lander environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lunar_lander.MLPDQN

Bases: Module

Neural network architecture for DQN compatible with Lunar Lander environment with the following parameters:

n_frames_stacked=1
append_step_count=False

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lunar_lander.MLPStepActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Lunar Lander environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lunar_lander.MLPStepCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Lunar Lander environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.lunar_lander.MLPStepDQN

Bases: Module

Neural network architecture for DQN compatible with Lunar Lander environment with the following parameters:

n_frames_stacked=1
append_step_count=True

forward(obs) → Tensor: Defines the computation performed at every call.

Models for MsPacman environment

See also

academia.environments.MsPacman

class academia.utils.models.ms_pacman.MLPActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Ms Pacman environment with the following parameters:

n_frames_stacked=1
append_step_count=False
obs_type='ram'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.ms_pacman.MLPCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Ms Pacman environment with the following parameters:

n_frames_stacked=1
append_step_count=False
obs_type='ram'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.ms_pacman.MLPDQN

Bases: Module

Neural network architecture for DQN compatible with Ms Pacman environment with the following parameters:

n_frames_stacked=1
append_step_count=False
obs_type='ram'

forward(x) → Tensor: Defines the computation performed at every call.

class academia.utils.models.ms_pacman.MLPStepActor

Bases: Module

Neural network architecture for PPO (actor) compatible with Ms Pacman environment with the following parameters:

n_frames_stacked=1
append_step_count=True
obs_type='ram'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.ms_pacman.MLPStepCritic

Bases: Module

Neural network architecture for PPO (critic) compatible with Ms Pacman environment with the following parameters:

n_frames_stacked=1
append_step_count=True
obs_type='ram'

forward(obs) → Tensor: Defines the computation performed at every call.

class academia.utils.models.ms_pacman.MLPStepDQN

Bases: Module

Neural network architecture for DQN compatible with Ms Pacman environment with the following parameters:

n_frames_stacked=1
append_step_count=True
obs_type='ram'

forward(x) → Tensor: Defines the computation performed at every call.