dialogue_manager.dialogue_policy.neural_dialogue_policy

Neural dialogue policy built on top of PyTorch.

Module Contents

Classes

NeuralDialoguePolicy

class dialogue_manager.dialogue_policy.neural_dialogue_policy.NeuralDialoguePolicy(input_size: int, hidden_size: int, output_size: int, possible_actions: List[Any])

Bases: torch.nn.Module

user_label_encoder

agent_label_encoder

abstract forward(state: torch.Tensor) → torch.Tensor

Forward pass of the policy.

Parameters:

state – State or batch of states.

Raises:

NotImplementedError – If the method is not implemented in the subclass.

Returns:

Output of the policy.

abstract select_action(state: torch.Tensor) → Any

Selects an action based on the current state.

Parameters:

state – The current state.

Raises:

NotImplementedError – If the method is not implemented in the subclass.

Returns:

Selected action and optionally other information.

abstract save_policy(path: str) → None

Saves the policy.

Parameters:

path – Path to save the policy.

Raises:

NotImplementedError – If the method is not implemented in the subclass.

abstract classmethod load_policy(path: str) → NeuralDialoguePolicy

Loads the policy.

Parameters:

path – Path to load the policy from.

Raises:

NotImplementedError – If the method is not implemented in the subclass.

Returns:

The loaded policy.

classmethod build_input_from_dialogue_state(dialogue_state: moviebot.dialogue_manager.dialogue_state.DialogueState, **kwargs) → torch.Tensor

Builds the input vector from the dialogue state.

The markovian state representation is built from booleans in the dialogue state (e.g., a recommendation was made, the agent should make an offer, we are at the beginning of the conversation). It can be seen as a one-hot encoding of the state.

Parameters:

dialogue_state – The dialogue state.

Returns:

Input vector for the policy (i.e., markovian state representation).

classmethod build_input_from_dialogue_state_and_intents(dialogue_state: moviebot.dialogue_manager.dialogue_state.DialogueState, user_intents: List[moviebot.core.intents.user_intents.UserIntents], agent_intents: List[moviebot.core.intents.agent_intents.AgentIntents], **kwargs) → torch.Tensor

Builds the input vector from the dialogue state and previous intents.

Parameters:

• dialogue_state – The dialogue state.

• user_intents – The user intents.

• agent_intents – The agent intents.

Returns:

The input vector.

classmethod build_input(dialogue_state: moviebot.dialogue_manager.dialogue_state.DialogueState, **kwargs) → torch.Tensor

Builds the input vector.

Parameters:

dialogue_state – The dialogue state.

Returns:

The input vector.