LinMeasurementLayerB¶

Role: Learnable mapping from a flattened gun vector to per-cell measurement probabilities in \([0, 1]\).

Location: Q_Sea_Battle.lin_measurement_layer_b.LinMeasurementLayerB

Derived constraints¶

Let n2 be the flattened field size (number of cells); n2 must be a positive int.
call() accepts rank-1 or rank-2 inputs only; the last dimension must equal n2 when statically known.

Constructor¶

Parameter	Type	Description
n2	int, constraint: `n2 > 0`, shape: scalar	Number of cells in the flattened representation; determines the output width of the final dense layer.
hidden_units	Sequence[int], constraint: each element convertible to `int`, shape: `(L,)`	Sizes of intermediate dense layers; each layer uses ReLU activation. Default: `(64,)`.
name	Optional[str], constraint: any valid Keras layer name or `None`, shape: scalar	Layer name passed to `tf.keras.layers.Layer`. Default: `"LinMeasurementLayerB"`.
**kwargs	dict[str, Unknown], constraint: passed through to `tf.keras.layers.Layer`, shape: mapping	Additional Keras layer keyword arguments.

Preconditions: n2 > 0.

Postconditions: self.n2 == int(n2); self.hidden_units is a tuple[int, ...]; internal MLP layer list is initialized empty and is constructed on first build().

Errors: Raises ValueError if n2 <= 0.

Example

import tensorflow as tf
from Q_Sea_Battle.lin_measurement_layer_b import LinMeasurementLayerB

n2 = 100
layer = LinMeasurementLayerB(n2=n2, hidden_units=(64, 32))

guns_batch = tf.random.uniform(shape=(8, n2), dtype=tf.float32)
probs_batch = layer(guns_batch, training=True)  # shape (8, n2)

guns_single = tf.random.uniform(shape=(n2,), dtype=tf.float32)
probs_single = layer(guns_single)  # shape (n2,)

Public Methods¶

build(input_shape) -> None¶

Create weights based on input shape.

Parameter input_shape: Unknown, constraint: Keras-compatible input shape descriptor, shape: Not specified.
Returns: None, constraint: not applicable, shape: not applicable.

Preconditions: None specified.

Postconditions: If not already built, appends len(hidden_units) hidden tf.keras.layers.Dense layers with ReLU activation and one output tf.keras.layers.Dense layer with units == n2 and sigmoid activation; sets internal built flag; calls super().build(input_shape).

Errors: Not specified.

call(guns, training: bool = False)¶

Forward pass.

Parameter guns: tf.Tensor-convertible, dtype: any (non-floating will be cast to tf.float32), shape: (B, n2) or (n2,).
Parameter training: bool, constraint: any boolean, shape: scalar.
Returns: tf.Tensor, dtype: floating (cast to tf.float32 if input is non-floating), constraint: elementwise in \([0, 1]\), shape: same as guns (returns (B, n2) for batched input and (n2,) for rank-1 input).

Preconditions: Input rank must be 1 or 2; last dimension must equal n2 when statically known.

Postconditions: Applies the internal MLP (Dense/ReLU layers followed by Dense/sigmoid) to produce probabilities; preserves original rank by temporarily expanding rank-1 inputs and squeezing the output back.

Errors: Raises ValueError if input rank is not 1 or 2; raises ValueError if the statically known last dimension is not n2.

Data & State¶

n2: int, constraint: n2 > 0, shape: scalar; output width and expected input last dimension.
hidden_units: tuple[int, ...], constraint: elements are int, shape: (L,); hidden layer widths.
_mlp: list[tf.keras.layers.Layer], constraint: contains Keras layers, shape: (L+1,) after build; internal sequence of Dense layers.
_built_mlp: bool, constraint: boolean, shape: scalar; indicates whether _mlp has been constructed.

Planned (design-spec)¶

Not specified.

Deviations¶