Skip to content

LinMeasurementLayerA

Role: Trainable Keras layer that maps rank-2 field logits to rank-2 measurement logits of the same width using a small MLP.

Location: Q_Sea_Battle.lin_measurement_layer_a.LinMeasurementLayerA

Constructor

Parameter | Type | Description | --- | --- | --- | | hidden_units | int, constraint \(\ge 1\), scalar | Number of units in the hidden Dense layer. | name | Optional[str], scalar | Optional layer name. | dtype | Optional[tf.dtypes.DType], scalar | Optional layer dtype; also used to set sublayer dtypes in build(). | **kwargs | Any, mapping | Additional keyword arguments forwarded to tf.keras.layers.Layer.

Preconditions - hidden_units is int-like and hidden_units >= 1.

Postconditions - self.hidden_units is set to int(hidden_units). - self._dense_hidden is None and self._dense_out is None until build(input_shape) is called by Keras (or manually).

Errors - ValueError: If hidden_units < 1.

Example 

import tensorflow as tf
from Q_Sea_Battle.lin_measurement_layer_a import LinMeasurementLayerA

layer = LinMeasurementLayerA(hidden_units=64, dtype=tf.float32)

x = tf.random.normal([8, 32])  # (B, n2)
y = layer(x, training=True)    # (B, n2)
print(y.shape)

Public Methods

build

Create sublayers once the input feature dimension is known.

Arguments - input_shape: Any, Keras shape-like, constraint: tf.TensorShape(input_shape)[-1] must be statically known (not None).

Returns - None, side-effect only.

Preconditions - The last dimension of input_shape is statically known; let \(n2 = \text{int}(\text{input\_shape}[-1])\).

Postconditions - self._dense_hidden is a tf.keras.layers.Dense, output dtype self.dtype, with units=self.hidden_units and activation="relu". - self._dense_out is a tf.keras.layers.Dense, output dtype self.dtype, with units=n2 and activation=None. - super().build(input_shape) has been called.

Errors - ValueError: If the last dimension of input_shape is None.

Example 

import tensorflow as tf
from Q_Sea_Battle.lin_measurement_layer_a import LinMeasurementLayerA

layer = LinMeasurementLayerA(hidden_units=16)
layer.build((None, 10))  # n2 = 10

call

Forward pass.

Arguments - field_batch: tf.Tensor, dtype float32 (or self.dtype if set), shape (B, n2); constraint: rank must be 2 when statically known. - training: bool, scalar; whether the call is in training mode. - **kwargs: Any, mapping; unused extra keyword arguments (kept for Keras compatibility).

Returns - tf.Tensor, dtype float32 (or self.dtype if set), shape (B, n2); measurement logits.

Preconditions - field_batch is convertible to a tensor via tf.convert_to_tensor. - If field_batch.shape.rank is statically known, it must equal 2. - The layer must have been built such that self._dense_hidden and self._dense_out are not None.

Postconditions - Computes h = Dense(hidden_units, relu)(field_batch) and returns Dense(n2, linear)(h).

Errors - ValueError: If field_batch has a statically-known rank and it is not 2. - RuntimeError: If self._dense_hidden is None or self._dense_out is None (layer not built correctly).

Example 

import tensorflow as tf
from Q_Sea_Battle.lin_measurement_layer_a import LinMeasurementLayerA

layer = LinMeasurementLayerA(hidden_units=32)
x = tf.random.normal([4, 12])  # (B, n2)
y = layer(x, training=False)

get_config

Return the serializable config for Keras.

Arguments - None.

Returns - Dict[str, Any], mapping; a Keras-serializable config including hidden_units.

Preconditions - None.

Postconditions - The returned dict equals super().get_config() updated with {"hidden_units": self.hidden_units}.

Errors - Not specified.

Example 

from Q_Sea_Battle.lin_measurement_layer_a import LinMeasurementLayerA

layer = LinMeasurementLayerA(hidden_units=8)
cfg = layer.get_config()
assert cfg["hidden_units"] == 8

Data & State

  • hidden_units: int, constraint \(\ge 1\), scalar; width of the hidden Dense layer.
  • _dense_hidden: Optional[tf.keras.layers.Dense], scalar reference; initialized to None in __init__, created in build().
  • _dense_out: Optional[tf.keras.layers.Dense], scalar reference; initialized to None in __init__, created in build().

Planned (design-spec)

  • Not specified.

Deviations

  • Not specified.

Notes for Contributors

  • _ensure_rank2(x, name) is a module-level helper used by call() and raises ValueError only when x.shape.rank is statically known and not equal to 2; it does not perform a dynamic (runtime) rank assertion.
  • call() forces field_batch through tf.convert_to_tensor(field_batch, dtype=self.dtype or tf.float32), so inputs may be cast to self.dtype (or float32 if self.dtype is unset).
  • tf.keras.layers.Layer
  • tf.keras.layers.Dense

Changelog

  • Not specified.