LinTrainableAssistedModelA¶

Role: Player A linear trainable-assisted baseline model that maps a field tensor to communication logits via measurement, assisted resource processing, and combination layers.

Location: Q_Sea_Battle.lin_trainable_assisted_model_a.LinTrainableAssistedModelA

Constructor¶

Parameter	Type	Description
field_size	int, constraint: convertible via `int(field_size)`, scalar	Field side length; used to derive \(n2 = field\_size \times field\_size\).
comms_size	int, constraint: convertible via `int(comms_size)`, scalar	Output communication size; forwarded to `LinCombineLayerA(comms_size=...)`.
sr_mode	str, constraint: any string, scalar	Shared resource mode; forwarded to `PRAssistedLayer(mode=...)`; default `"expected"`.
seed	int \| None, constraint: any integer or `None`, scalar	Seed forwarded to `PRAssistedLayer(seed=...)`; default `0`.
p_high	float, constraint: convertible via `float(p_high)`, scalar	High probability forwarded to `PRAssistedLayer(p_high=...)`; default `0.9`.
resource_index	int, constraint: convertible via `int(resource_index)`, scalar	Resource index forwarded to `PRAssistedLayer(resource_index=...)`; default `0`.
hidden_units_meas	Sequence[int], constraint: sequence of ints, shape (m,)	Hidden units configuration forwarded to `LinMeasurementLayerA(hidden_units=...)`; default `(64,)`.
hidden_units_combine	int \| Sequence[int], constraint: int or sequence of ints, scalar or shape (m,)	Hidden units configuration forwarded to `LinCombineLayerA(hidden_units=...)`; default `(64, 64)`.
name	str \| None, constraint: any string or `None`, scalar	Keras model name; if `None`, uses `"LinTrainableAssistedModelA"`.
**kwargs	Any, constraint: arbitrary keyword args	Forwarded to `tf.keras.Model.__init__`.

Preconditions: field_size and comms_size are provided and convertible to int; field_size should be meaningful for forming \(n2 = field\_size^2\) (not otherwise validated in code); field_batch inputs provided later should be convertible to tf.Tensor of dtype float32 in _ensure_batched.
Postconditions: Attributes are created: field_size (int, scalar), comms_size (int, scalar), n2 (int, scalar), measurement (LinMeasurementLayerA), pr_assisted (PRAssistedLayer), combine (LinCombineLayerA), plus backward-compatible aliases measure_layer, sr_layer, combine_layer pointing to the same objects.
Errors: Not specified; any exceptions raised by TensorFlow/Keras or the referenced layers’ constructors may propagate.

Example

import tensorflow as tf
from Q_Sea_Battle.lin_trainable_assisted_model_a import LinTrainableAssistedModelA

model = LinTrainableAssistedModelA(field_size=5, comms_size=8)
x = tf.zeros((2, 25), dtype=tf.float32)
y = model(x, training=False)

Public Methods¶

_ensure_batched¶

Signature: _ensure_batched(x: tf.Tensor) -> Tuple[tf.Tensor, bool] (static method)

Arguments: - x: tf.Tensor, dtype: any convertible to float32 via tf.convert_to_tensor(..., dtype=tf.float32), shape (n2,) or (B, n2) where B is batch size

Returns: - batched_x: tf.Tensor, dtype float32, shape (1, n2) if input rank is 1 else shape (B, n2) - was_batched: bool, constraint: False if input rank is 1 else True, scalar

Behavior: Converts x to a float32 tensor; if x.shape.rank == 1, expands a batch dimension at axis 0 and returns (expanded, False), otherwise returns (x, True).
Errors: Not specified; conversion/shape rank access errors may propagate from TensorFlow.

compute_with_internal¶

Signature: compute_with_internal(self, field_batch: tf.Tensor, training: bool = False) -> Tuple[tf.Tensor, List[tf.Tensor], List[tf.Tensor]]

Arguments: - field_batch: tf.Tensor, dtype: any convertible to float32, shape (n2,) or (B, n2) - training: bool, constraint: boolean, scalar

Returns: - comm_logits: tf.Tensor, dtype: Not specified (produced by LinCombineLayerA), shape (B, comms_size) if the combine layer follows the typical convention (exact shape not specified in this module) - measurements: List[tf.Tensor], length 1; element 0 is meas_for_resource: tf.Tensor, dtype float32, shape (B, n2) - internals: List[tf.Tensor], length 1; element 0 is outcomes: tf.Tensor, dtype: Not specified (produced by PRAssistedLayer), shape: Not specified

Behavior: Ensures field_batch is batched float32; computes measurement probabilities via self.measurement(field_batch, training=training); if getattr(self.pr_assisted, "mode", "expected") == "sample", binarizes measurements with threshold >= 0.5 (float32), otherwise uses probabilities as-is; calls self.pr_assisted with a dict containing current_measurement, zero tensors for previous_measurement and previous_outcome, and first_measurement set to ones of shape (B, 1); combines assisted outcomes via self.combine(outcomes, training=training); returns logits plus internal tensors wrapped in lists.
Errors: Not specified; any exceptions raised by the underlying layers or TensorFlow ops may propagate.

call¶

Signature: call(self, field_batch: tf.Tensor, training: bool = False) -> tf.Tensor

Arguments: - field_batch: tf.Tensor, dtype: any convertible to float32, shape (n2,) or (B, n2) - training: bool, constraint: boolean, scalar

Returns: - comm_logits: tf.Tensor, dtype: Not specified (produced by LinCombineLayerA), shape: Not specified in this module

Behavior: Delegates to compute_with_internal(..., training=training) and returns only comm_logits.
Errors: Not specified; propagates errors from compute_with_internal.

Data & State¶

field_size: int, constraint: int(field_size) conversion applied, scalar; used to derive n2.
comms_size: int, constraint: int(comms_size) conversion applied, scalar.
n2: int, constraint: equals field_size * field_size, scalar; flattened field length.
measurement: LinMeasurementLayerA, constraint: constructed as LinMeasurementLayerA(n2=self.n2, hidden_units=hidden_units_meas).
pr_assisted: PRAssistedLayer, constraint: constructed as PRAssistedLayer(length=self.n2, p_high=float(p_high), mode=str(sr_mode), resource_index=int(resource_index), seed=seed, name="PRAssistedLayerA").
combine: LinCombineLayerA, constraint: constructed as LinCombineLayerA(comms_size=self.comms_size, hidden_units=hidden_units_combine).
measure_layer: LinMeasurementLayerA, alias of measurement (backward-compatible).
sr_layer: PRAssistedLayer, alias of pr_assisted (backward-compatible).
combine_layer: LinCombineLayerA, alias of combine (backward-compatible).

Planned (design-spec)¶

Not specified.

Deviations¶