TrainableAssistedPlayers

Role: Players-style wrapper that wires a trainable sender (A) and receiver (B) assisted player pair and owns their underlying models and shared state.

Location: Q_Sea_Battle.trainable_assisted_players.TrainableAssistedPlayers

Constructor

Parameter	Type	Description
game_layout	Any, not specified, must provide attributes field_size and comms_size convertible to int	Game-layout-like object used to parameterize default models and passed into player wrappers.
p_high	float, not specified	Forward-compatibility parameter; currently unused by this class when building default linear models.
num_iterations	Optional[int], not specified	Forward-compatibility parameter; currently unused by this class when building default linear models.
hidden_dim	int, not specified	Forward-compatibility parameter; currently unused by this class when building default linear models.
L_meas	Optional[int], not specified	Forward-compatibility parameter; currently unused by this class when building default linear models.
model_a	Optional[LinTrainableAssistedModelA], default None	If provided, used as the A-side model; otherwise a default LinTrainableAssistedModelA is constructed from game_layout.field_size and game_layout.comms_size.
model_b	Optional[LinTrainableAssistedModelB], default None	If provided, used as the B-side model; otherwise a default LinTrainableAssistedModelB is constructed from game_layout.field_size and game_layout.comms_size.

Preconditions

• game_layout has attributes field_size and comms_size such that int(getattr(game_layout, "field_size")) and int(getattr(game_layout, "comms_size")) succeed when default models are constructed.

Postconditions

• self.game_layout is set to game_layout (type: Any, constraints: not specified).
• self.explore is initialized to False (type: bool, constraints: {True, False}).
• self.model_a is set (type: LinTrainableAssistedModelA, constraints: not specified).
• self.model_b is set (type: LinTrainableAssistedModelB, constraints: not specified).
• self.previous is initialized to None (type: Any | None, constraints: typically either None or a tuple (measurements_per_layer, outcomes_per_layer) where both are Python lists of tensors each shaped (B, n2)).
• Lazy player wrappers self._playerA and self._playerB are initialized to None (type: Optional[TrainableAssistedPlayerA] and Optional[TrainableAssistedPlayerB], constraints: not specified).
• self.has_prev is initialized to True (type: bool, constraints: {True, False}).

Errors

• Any exception raised by int(getattr(game_layout, "field_size")) or int(getattr(game_layout, "comms_size")) when constructing default models.
• Any exception raised by LinTrainableAssistedModelA(...) or LinTrainableAssistedModelB(...) constructors.

Example

from Q_Sea_Battle.trainable_assisted_players import TrainableAssistedPlayers

class Layout:
    field_size = 10
    comms_size = 2

tap = TrainableAssistedPlayers(Layout())
player_a, player_b = tap.players()
tap.set_explore(True)
tap.reset()

Public Methods

check_model_correspondence

Check that model A and B are compatible by comparing exposed field_size and comms_size attributes if present.

Arguments

• None

Returns

• bool, constraints: {True, False}, shape: scalar.

Errors

• Not specified; internal exceptions while accessing model attributes are caught and result in True.

players

Return the (PlayerA, PlayerB) wrappers, constructing them lazily and persisting wrapper state until reset().

Arguments

• None

Returns

• Tuple[TrainableAssistedPlayerA, TrainableAssistedPlayerB], constraints: 2-tuple (player_a, player_b), shape: length 2.

Errors

• Any exception raised by TrainableAssistedPlayerA(self.game_layout, model_a=self.model_a) or TrainableAssistedPlayerB(self.game_layout, model_b=self.model_b) during lazy construction.

reset

Reset internal state between games by resetting both wrappers (if they exist) and clearing previous.

Arguments

• None

Returns

• NoneType, constraints: always None, shape: scalar.

Errors

• Any exception raised by self._playerA.reset() or self._playerB.reset() if those wrappers exist.

set_explore

Set the exploration flag for both players and the wrapper itself.

Arguments

• flag: bool, constraints: {True, False}, shape: scalar.

Returns

• NoneType, constraints: always None, shape: scalar.

Errors

• Not specified.

Data & State

• has_log_probs: bool, constraints: {True, False}, shape: scalar; class attribute set to True.
• game_layout: Any, constraints: not specified; expected to provide field_size and comms_size attributes.
• model_a: LinTrainableAssistedModelA, constraints: not specified.
• model_b: LinTrainableAssistedModelB, constraints: not specified.
• explore: bool, constraints: {True, False}, shape: scalar; shared exploration flag propagated to wrappers when they exist.
• previous: Any | None, constraints: typically (measurements_per_layer, outcomes_per_layer) or None; when present, expected contract is (meas_list, out_list) where both are Python lists of tensors each shaped (B, n2).
• has_prev: bool, constraints: {True, False}, shape: scalar; initialized to True.
• _playerA: Optional[TrainableAssistedPlayerA], constraints: either None or an instantiated wrapper; created lazily by players().
• _playerB: Optional[TrainableAssistedPlayerB], constraints: either None or an instantiated wrapper; created lazily by players().

Planned (design-spec)

• Not specified.

Deviations

• Not specified.

Notes for Contributors

• p_high, num_iterations, hidden_dim, and L_meas are accepted by the constructor for forward compatibility but are currently unused by this class when creating default linear models.
• The previous state is intended to be produced by Player A and consumed by Player B via the parent reference set in players(); this module does not enforce tensor types beyond the documented contract in the module docstring.

Related

• Q_Sea_Battle.trainable_assisted_player_a.TrainableAssistedPlayerA
• Q_Sea_Battle.trainable_assisted_player_b.TrainableAssistedPlayerB
• Q_Sea_Battle.lin_trainable_assisted_model_a.LinTrainableAssistedModelA
• Q_Sea_Battle.lin_trainable_assisted_model_b.LinTrainableAssistedModelB

Changelog

• 0.1: Initial version (as indicated by module docstring).