LinTrainableAssistedModelB¶

Role: Backward-compatible wrapper that preserves a legacy constructor interface while delegating learning behavior to LinInternalModelB.

Location: Q_Sea_Battle.lin_trainable_assisted_model_b.LinTrainableAssistedModelB

Constructor¶

Parameter	Type	Description
field_size	int, constraint: convertible via `int(field_size)`; shape: scalar	Game field size used to construct the internal layout; stored as `self.field_size`.
comms_size	int, constraint: convertible via `int(comms_size)`; shape: scalar	Communication channel size used to construct the internal layout; stored as `self.comms_size`.
sr_mode	str, constraint: must map via `_map_sr_mode` to one of `{"stochastic","replay"}`; shape: scalar	Shared resource mode; legacy aliases accepted: `"sample"` maps to `"stochastic"`, `"expected"` maps to `"replay"`; default `"sample"`.
p_rule	float, constraint: not specified; shape: scalar	Probability of applying the rule-based component; forwarded to the internal model; default `1.0`.
beta	float, constraint: not specified; shape: scalar	Hyperparameter forwarded to the internal model; default `10.0`.
alpha	float, constraint: not specified; shape: scalar	Hyperparameter forwarded to the internal model; default `5.0`.
seed	Optional[int], constraint: `None` or int; shape: scalar	Optional RNG seed forwarded to the internal model; default `None`.
hidden_units_measure	int, constraint: not specified; shape: scalar	Width of the measurement subnetwork forwarded to the internal model; default `64`.
hidden_units_combine	int, constraint: not specified; shape: scalar	Width of the combine subnetwork forwarded to the internal model; default `64`.
name	Optional[str], constraint: `None` or str; shape: scalar	Optional model name forwarded to the internal model; default `None`.
**kwargs	Any, constraint: accepted but unused by this wrapper; shape: N/A	Extra keyword arguments accepted for backward compatibility; currently unused.

Preconditions

field_size and comms_size are convertible to int via int(...).
sr_mode is convertible to str and lowercased value is one of: "sample", "stochastic", "expected", "replay".

Postconditions

self.field_size is set to int(field_size) and self.comms_size is set to int(comms_size).
The base class initializer LinInternalModelB.__init__ has been invoked with an ad-hoc layout object providing layout.field_size and layout.comms_size, and with sr_mode normalized to "stochastic" or "replay".

Errors

Raises ValueError if sr_mode is not a supported alias/value (as determined by _map_sr_mode).

Example

Construct using legacy SR mode alias

from Q_Sea_Battle.lin_trainable_assisted_model_b import LinTrainableAssistedModelB

model = LinTrainableAssistedModelB(
    field_size=10,
    comms_size=4,
    sr_mode="sample",  # legacy alias; normalized to "stochastic"
    seed=123,
    name="player_model",
)

Public Methods¶

Not specified in this module. This class inherits all public methods from LinInternalModelB.

Data & State¶

field_size: int, constraint: set by int(field_size); shape: scalar; legacy field size retained on the wrapper instance.
comms_size: int, constraint: set by int(comms_size); shape: scalar; legacy comms size retained on the wrapper instance.

Other data/state are inherited from LinInternalModelB (not specified in this module).

Planned (design-spec)¶

Not specified.

Deviations¶

Not specified.

Notes for Contributors¶

This wrapper exists to preserve a historical constructor signature while providing a minimal layout object containing only field_size and comms_size.
The internal model interface and learning logic are intentionally not duplicated here; changes to learning behavior should be made in LinInternalModelB.
Legacy SR mode normalization is performed by the module-private _map_sr_mode helper; if adding new SR modes, ensure aliases remain backward compatible.

Q_Sea_Battle.lin_internal_model_b.LinInternalModelB (base class; actual implementation target).

Changelog¶

Not specified.