fusionlab.nn.components.AnomalyLoss

class fusionlab.nn.components.AnomalyLoss

Bases: Loss, NNLearner

Anomaly Loss layer computing mean squared anomaly scores.

This layer expects anomaly scores of shape \((B, H, D)\) and multiplies their mean squared value by a weight factor.

\[\text{AnomalyLoss}(\mathbf{a}) = w \cdot \frac{1}{BHD} \sum (\mathbf{a})^2\]

where \(\mathbf{a}\) is the anomaly score, and \(w\) is the weight.

Parameters:

weight (float, optional) – Scalar multiplier for the computed mean squared anomaly scores. Defaults to 1.0.

Notes

Anomaly loss is often combined with other losses in a multi-task setting where predictive performance and anomaly detection performance are both important.

call(`anomaly_scores`)

Compute mean squared anomaly loss.

Parameters:

anomaly_scores (Tensor)

get_config()

Return configuration for serialization.

from_config(`config`)

Instantiates a new instance from config.

Examples

>>> from fusionlab.nn.components import AnomalyLoss
>>> import tensorflow as tf
>>> # Suppose anomaly_scores is (B, H, D)
>>> anomaly_scores = tf.random.normal((32, 10, 8))
>>> # Instantiate anomaly loss
>>> anomaly_loss_fn = AnomalyLoss(weight=2.0)
>>> # Compute anomaly loss
>>> loss_value = anomaly_loss_fn(anomaly_scores)

See also

AdaptiveQuantileLoss

Another specialized loss that can be combined for multi-objective optimization.

MultiObjectiveLoss

Demonstrates how anomaly and quantile loss can be merged.

References

[1]

Lim, B., & Zohren, S. (2021). “Time-series forecasting with deep learning: a survey.” Philosophical Transactions of the Royal Society A, 379(2194), 20200209.

__init__(weight=1.0, name='AnomalyLoss')

Initializes Loss class.

Parameters:

• reduction – Type of tf.keras.losses.Reduction to apply to loss. Default value is AUTO. AUTO indicates that the reduction option will be determined by the usage context. For almost all cases this defaults to SUM_OVER_BATCH_SIZE. When used under a tf.distribute.Strategy, except via Model.compile() and Model.fit(), using AUTO or SUM_OVER_BATCH_SIZE will raise an error. Please see this custom training [tutorial]( https://www.tensorflow.org/tutorials/distribute/custom_training) for more details.

• name – Optional name for the instance.

• weight (float)

Methods

__init__([weight, name])	Initializes Loss class.
call(anomaly_scores[, y_pred])	Forward pass that computes the mean squared anomaly score multiplied by weight.
from_config(config)	Recreates an AnomalyLoss layer from a config.
get_config()	Return configuration dictionary for this layer.
get_params([deep])	Get the parameters for this learner.
help(**kwargs)
load(file_path[, format])	Load the learner's state from a specified file in the desired format.
save([file_path, format, overwrite, ...])	Save the learner's state to a specified file in the desired format.
set_params(**params)	Set the parameters of this learner.
summary()	Provide a summary of the learner's parameters.

Attributes

my_params

__init__(weight=1.0, name='AnomalyLoss')

Initializes Loss class.

Parameters:

• reduction – Type of tf.keras.losses.Reduction to apply to loss. Default value is AUTO. AUTO indicates that the reduction option will be determined by the usage context. For almost all cases this defaults to SUM_OVER_BATCH_SIZE. When used under a tf.distribute.Strategy, except via Model.compile() and Model.fit(), using AUTO or SUM_OVER_BATCH_SIZE will raise an error. Please see this custom training [tutorial]( https://www.tensorflow.org/tutorials/distribute/custom_training) for more details.

• name – Optional name for the instance.

• weight (float)

call(anomaly_scores, y_pred=None)

Forward pass that computes the mean squared anomaly score multiplied by weight.

Parameters:

• anomaly_scores (tf.Tensor) – Tensor of shape (B, H, D) representing anomaly scores.

• y_pred (Optional) – Does nothing, just for API consistency.

Returns:

A scalar loss value representing the weighted mean squared anomaly.

Return type:

tf.Tensor

get_config()

Return configuration dictionary for this layer.

Returns:

Includes ‘weight’.

Return type:

dict

classmethod from_config(config)

Recreates an AnomalyLoss layer from a config.

Parameters:

config (dict) – Configuration containing ‘weight’.

Returns:

A new instance of this layer.

Return type:

AnomalyLoss

help(**kwargs)

my_params = AnomalyLoss(weight=1.0, name='AnomalyLoss')