nip.scenario_instance.ScenarioInstance#
- class nip.scenario_instance.ScenarioInstance(train_dataset: Dataset, test_dataset: Dataset, agents: dict[str, Agent], protocol_handler: ProtocolHandler, message_regressor: MessageRegressor, train_environment: Environment | None = None, test_environment: Environment | None = None, combined_whole: CombinedWhole | None = None, combined_body: CombinedBody | None = None, combined_policy_body: CombinedBody | None = None, combined_value_body: CombinedBody | None = None, combined_policy_head: CombinedPolicyHead | None = None, combined_value_head: CombinedValueHead | None = None, shared_model_groups: dict[str, PureTextSharedModelGroup] | None = None)[source]#
A dataclass for holding the components of an experiment.
The principal aim of this class is to abstract away the details of the particular experiment being run.
- test_dataset#
The dataset used for testing. Note that this can be either the “test” or “validation” split, depending on the hyperparameters.
- Type:
- protocol_handler#
The interaction protocol handler for the experiment.
- Type:
- message_regressor#
The message regressor for the experiment, which is used to test if the label can be inferred purely from the messages.
- Type:
MessageRegressor
- agents#
The agents for the experiment. Each ‘agent’ is a dictionary containing all of the agent parts.
- train_environment#
The train environment for the experiment, if the experiment is RL.
- Type:
Optional[Environment]
- test_environment#
The environment for testing the agents, which uses
test_dataset. Note that this can be either the “test” or “validation” split, depending on the hyperparameters.- Type:
Optional[Environment]
- combined_whole#
If the agents are not split into parts, this holds the combination of the whole agents.
- Type:
Optional[CombinedWholeAgent]
- combined_body#
The combined body of the agents, if the agents are combined the actor and critic share the same body.
- Type:
Optional[CombinedBody]
- combined_policy_body#
The combined policy body of the agents, if the agents are combined and the actor and critic have separate bodies.
- Type:
Optional[CombinedBody]
- combined_value_body#
The combined value body of the agents, if the agents are combined and the actor and critic have separate bodies.
- Type:
Optional[CombinedBody]
- combined_policy_head#
The combined policy head of the agents, if the agents are combined.
- Type:
Optional[CombinedPolicyHead]
- combined_value_head#
The combined value head of the agents, if the agents are combined.
- Type:
Optional[CombinedValueHead]
The shared model groups for pure-text environments. Agents in the same group share the same model. A dictionary with the group name as the key and the shared model group as the value.
- Type:
Optional[dict[str, PureTextSharedModelGroup]]
Methods Summary
__eq__(other)Return self==value.
__init__(train_dataset, test_dataset, ...[, ...])__repr__()Return repr(self).
Attributes
Methods
- __eq__(other)#
Return self==value.
- __init__(train_dataset: Dataset, test_dataset: Dataset, agents: dict[str, Agent], protocol_handler: ProtocolHandler, message_regressor: MessageRegressor, train_environment: Environment | None = None, test_environment: Environment | None = None, combined_whole: CombinedWhole | None = None, combined_body: CombinedBody | None = None, combined_policy_body: CombinedBody | None = None, combined_value_body: CombinedBody | None = None, combined_policy_head: CombinedPolicyHead | None = None, combined_value_head: CombinedValueHead | None = None, shared_model_groups: dict[str, PureTextSharedModelGroup] | None = None) None#
- __repr__()#
Return repr(self).