alibi_detect.models.tensorflow.losses

Functions

elbo

elbo(y_true: tensorflow.python.framework.tensor.Tensor, y_pred: tensorflow.python.framework.tensor.Tensor, cov_full: Optional[tensorflow.python.framework.tensor.Tensor] = None, cov_diag: Optional[tensorflow.python.framework.tensor.Tensor] = None, sim: Optional[float] = None) -> tensorflow.python.framework.tensor.Tensor

Compute ELBO loss. The covariance matrix can be specified by passing the full covariance matrix, the matrix

diagonal, or a scale identity multiplier. Only one of these should be specified. If none are specified, the identity matrix is used.

Example

import tensorflow as tf from alibi_detect.models.tensorflow.losses import elbo y_true = tf.constant([[0.0, 1.0], [1.0, 0.0]]) y_pred = tf.constant([[0.1, 0.9], [0.8, 0.2]])

Specifying scale identity multiplier

elbo(y_true, y_pred, sim=1.0)

Specifying covariance matrix diagonal

elbo(y_true, y_pred, cov_diag=tf.ones(2))

Specifying full covariance matrix

elbo(y_true, y_pred, cov_full=tf.eye(2))

Name	Type	Default	Description
y_true	tensorflow.python.framework.tensor.Tensor		Labels.
y_pred	tensorflow.python.framework.tensor.Tensor		Predictions.
cov_full	Optional[tensorflow.python.framework.tensor.Tensor]	None	Full covariance matrix.
cov_diag	Optional[tensorflow.python.framework.tensor.Tensor]	None	Diagonal (variance) of covariance matrix.
sim	Optional[float]	None	Scale identity multiplier.

Returns

• Type: tensorflow.python.framework.tensor.Tensor

loss_adv_ae

loss_adv_ae(x_true: tensorflow.python.framework.tensor.Tensor, x_pred: tensorflow.python.framework.tensor.Tensor, model: Optional[keras.src.models.model.Model] = None, model_hl: Optional[list] = None, w_model: float = 1.0, w_recon: float = 0.0, w_model_hl: Optional[list] = None, temperature: float = 1.0) -> tensorflow.python.framework.tensor.Tensor

Loss function used for AdversarialAE.

Name	Type	Default	Description
x_true	tensorflow.python.framework.tensor.Tensor		Batch of instances.
x_pred	tensorflow.python.framework.tensor.Tensor		Batch of reconstructed instances by the autoencoder.
model	Optional[keras.src.models.model.Model]	None	A trained tf.keras model with frozen layers (layers.trainable = False).
model_hl	Optional[list]	None	List with tf.keras models used to extract feature maps and make predictions on hidden layers.
w_model	float	1.0	Weight on model prediction loss term.
w_recon	float	0.0	Weight on MSE reconstruction error loss term.
w_model_hl	Optional[list]	None	Weights assigned to the loss of each model in model_hl.
temperature	float	1.0	Temperature used for model prediction scaling. Temperature <1 sharpens the prediction probability distribution.

Returns

• Type: tensorflow.python.framework.tensor.Tensor

loss_aegmm

loss_aegmm(x_true: tensorflow.python.framework.tensor.Tensor, x_pred: tensorflow.python.framework.tensor.Tensor, z: tensorflow.python.framework.tensor.Tensor, gamma: tensorflow.python.framework.tensor.Tensor, w_energy: float = 0.1, w_cov_diag: float = 0.005) -> tensorflow.python.framework.tensor.Tensor

Loss function used for OutlierAEGMM.

Name	Type	Default	Description
x_true	tensorflow.python.framework.tensor.Tensor		Batch of instances.
x_pred	tensorflow.python.framework.tensor.Tensor		Batch of reconstructed instances by the autoencoder.
z	tensorflow.python.framework.tensor.Tensor		Latent space values.
gamma	tensorflow.python.framework.tensor.Tensor		Membership prediction for mixture model components.
w_energy	float	0.1	Weight on sample energy loss term.
w_cov_diag	float	0.005	Weight on covariance regularizing loss term.

Returns

• Type: tensorflow.python.framework.tensor.Tensor

loss_distillation

loss_distillation(x_true: tensorflow.python.framework.tensor.Tensor, y_pred: tensorflow.python.framework.tensor.Tensor, model: Optional[keras.src.models.model.Model] = None, loss_type: str = 'kld', temperature: float = 1.0) -> tensorflow.python.framework.tensor.Tensor

Loss function used for Model Distillation.

Name	Type	Default	Description
x_true	tensorflow.python.framework.tensor.Tensor		Batch of data points.
y_pred	tensorflow.python.framework.tensor.Tensor		Batch of prediction from the distilled model.
model	Optional[keras.src.models.model.Model]	None	tf.keras model.
loss_type	str	'kld'	Type of loss for distillation. Supported 'kld', 'xent.
temperature	float	1.0	Temperature used for model prediction scaling. Temperature <1 sharpens the prediction probability distribution.

Returns

• Type: tensorflow.python.framework.tensor.Tensor

loss_vaegmm

loss_vaegmm(x_true: tensorflow.python.framework.tensor.Tensor, x_pred: tensorflow.python.framework.tensor.Tensor, z: tensorflow.python.framework.tensor.Tensor, gamma: tensorflow.python.framework.tensor.Tensor, w_recon: float = 1e-07, w_energy: float = 0.1, w_cov_diag: float = 0.005, cov_full: Optional[tensorflow.python.framework.tensor.Tensor] = None, cov_diag: Optional[tensorflow.python.framework.tensor.Tensor] = None, sim: float = 0.05) -> tensorflow.python.framework.tensor.Tensor

Loss function used for OutlierVAEGMM.

Name	Type	Default	Description
x_true	tensorflow.python.framework.tensor.Tensor		Batch of instances.
x_pred	tensorflow.python.framework.tensor.Tensor		Batch of reconstructed instances by the variational autoencoder.
z	tensorflow.python.framework.tensor.Tensor		Latent space values.
gamma	tensorflow.python.framework.tensor.Tensor		Membership prediction for mixture model components.
w_recon	float	1e-07	Weight on elbo loss term.
w_energy	float	0.1	Weight on sample energy loss term.
w_cov_diag	float	0.005	Weight on covariance regularizing loss term.
cov_full	Optional[tensorflow.python.framework.tensor.Tensor]	None	Full covariance matrix.
cov_diag	Optional[tensorflow.python.framework.tensor.Tensor]	None	Diagonal (variance) of covariance matrix.
sim	float	0.05	Scale identity multiplier.

Returns

• Type: tensorflow.python.framework.tensor.Tensor