fusionlab.nn.components.ExplainableAttention

class fusionlab.nn.components.ExplainableAttention

Bases: Layer, NNLearner

ExplainableAttention layer that returns attention scores from multi-head attention [1].

This layer is useful for interpretability, providing insight into how the attention mechanism focuses on different time steps.

\[\mathbf{A} = \text{MHA}(\mathbf{X},\,\mathbf{X}) \rightarrow \text{attention\_scores}\]

Here, \(\mathbf{X}\) is an input tensor, and attention_scores is the matrix capturing attention weights.

Parameters:

• num_heads (int) – Number of heads for multi-head attention.

• key_dim (int) – Dimensionality of the query/key projections.

Notes

Unlike standard layers that return the transformation output, this layer specifically returns the attention score matrix for interpretability.

call(`inputs`, training=False)

Forward pass that outputs only the attention scores.

get_config()

Returns the configuration for serialization.

from_config(`config`)

Creates a new instance from the given config.

Examples

>>> from fusionlab.nn.components import ExplainableAttention
>>> import tensorflow as tf
>>> # Suppose we have input of shape (batch_size, time_steps, features)
>>> x = tf.random.normal((32, 10, 64))
>>> # Instantiate explainable attention
>>> ea = ExplainableAttention(num_heads=4, key_dim=64)
>>> # Forward pass returns attention scores: (B, num_heads, T, T)
>>> scores = ea(x)

See also

CrossAttention

Another attention variant for cross-sequence contexts.

MultiResolutionAttentionFusion

For fusing features via multi-head attention.

References

[1]

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017). “Attention is all you need.” In Advances in Neural Information Processing Systems (pp. 5998-6008).

__init__(num_heads, key_dim)

Initialize the ExplainableAttention layer.

Parameters:

• num_heads (int) – Number of attention heads.

• key_dim (int) – Dimensionality of query/key projections in multi-head attention.

Methods

__init__(num_heads, key_dim)	Initialize the ExplainableAttention layer.
add_loss(loss)	Can be called inside of the call() method to add a scalar loss.
add_metric(*args, **kwargs)
add_variable(shape, initializer[, dtype, ...])	Add a weight variable to the layer.
add_weight([shape, initializer, dtype, ...])	Add a weight variable to the layer.
build(input_shape)
build_from_config(config)	Builds the layer's states with the supplied config dict.
call(inputs[, training])	Forward pass that returns only the attention scores.
compute_mask(inputs, previous_mask)
compute_output_shape(*args, **kwargs)
compute_output_spec(*args, **kwargs)
count_params()	Count the total number of scalars composing the weights.
from_config(config)	Creates a new instance from the config dictionary.
get_build_config()	Returns a dictionary with the layer's input shape.
get_config()	Returns the layer configuration.
get_params([deep])	Get the parameters for this learner.
get_weights()	Return the values of layer.weights as a list of NumPy arrays.
help(**kwargs)
load(file_path[, format])	Load the learner's state from a specified file in the desired format.
load_own_variables(store)	Loads the state of the layer.
quantize(mode[, type_check, config])
quantized_build(input_shape, mode)
quantized_call(*args, **kwargs)
rematerialized_call(layer_call, *args, **kwargs)	Enable rematerialization dynamically for layer's call method.
save([file_path, format, overwrite, ...])	Save the learner's state to a specified file in the desired format.
save_own_variables(store)	Saves the state of the layer.
set_params(**params)	Set the parameters of this learner.
set_weights(weights)	Sets the values of layer.weights from a list of NumPy arrays.
stateless_call(trainable_variables, ...[, ...])	Call the layer without any side effects.
summary()	Provide a summary of the learner's parameters.
symbolic_call(*args, **kwargs)

Attributes

compute_dtype	The dtype of the computations performed by the layer.
dtype	Alias of layer.variable_dtype.
dtype_policy
input	Retrieves the input tensor(s) of a symbolic operation.
input_dtype	The dtype layer inputs should be converted to.
input_spec
losses	List of scalar losses from add_loss, regularizers and sublayers.
metrics	List of all metrics.
metrics_variables	List of all metric variables.
my_params
non_trainable_variables	List of all non-trainable layer state.
non_trainable_weights	List of all non-trainable weight variables of the layer.
output	Retrieves the output tensor(s) of a layer.
path	The path of the layer.
quantization_mode	The quantization mode of this layer, None if not quantized.
supports_masking	Whether this layer supports computing a mask using compute_mask.
trainable	Settable boolean, whether this layer should be trainable or not.
trainable_variables	List of all trainable layer state.
trainable_weights	List of all trainable weight variables of the layer.
variable_dtype	The dtype of the state (weights) of the layer.
variables	List of all layer state, including random seeds.
weights	List of all weight variables of the layer.

__init__(num_heads, key_dim)

Initialize the ExplainableAttention layer.

Parameters:

• num_heads (int) – Number of attention heads.

• key_dim (int) – Dimensionality of query/key projections in multi-head attention.

call(inputs, training=False)

Forward pass that returns only the attention scores.

Parameters:

• inputs (tf.Tensor) – Tensor of shape (B, T, D).

• training (bool, optional) – Indicates training mode; not used in this layer. Defaults to False.

Returns:

Attention scores of shape (B, num_heads, T, T).

Return type:

tf.Tensor

get_config()

Returns the layer configuration.

Returns:

Dictionary containing ‘num_heads’ and ‘key_dim’.

Return type:

dict

classmethod from_config(config)

Creates a new instance from the config dictionary.

Parameters:

config (dict) – Configuration dictionary.

Returns:

A new instance of this layer.

Return type:

ExplainableAttention

help(**kwargs)

my_params = ExplainableAttention(num_heads, key_dim)