alibi_detect.models.tensorflow.autoencoder

AE

Inherits from: Model, TensorFlowTrainer, Trainer, Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

AE(self, encoder_net: keras.src.models.model.Model, decoder_net: keras.src.models.model.Model, name: str = 'ae') -> None

Name	Type	Default	Description
encoder_net	keras.src.models.model.Model		Layers for the encoder wrapped in a tf.keras.Sequential class.
decoder_net	keras.src.models.model.Model		Layers for the decoder wrapped in a tf.keras.Sequential class.
name	str	'ae'	Name of autoencoder model.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> tensorflow.python.framework.tensor.Tensor

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

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

AEGMM

Inherits from: Model, TensorFlowTrainer, Trainer, Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

AEGMM(self, encoder_net: keras.src.models.model.Model, decoder_net: keras.src.models.model.Model, gmm_density_net: keras.src.models.model.Model, n_gmm: int, recon_features: Callable = <function eucl_cosim_features at 0x280c66430>, name: str = 'aegmm') -> None

Name	Type	Default	Description
encoder_net	keras.src.models.model.Model		Layers for the encoder wrapped in a tf.keras.Sequential class.
decoder_net	keras.src.models.model.Model		Layers for the decoder wrapped in a tf.keras.Sequential class.
gmm_density_net	keras.src.models.model.Model		Layers for the GMM network wrapped in a tf.keras.Sequential class.
n_gmm	int		Number of components in GMM.
recon_features	Callable	<function eucl_cosim_features at 0x280c66430>	Function to extract features from the reconstructed instance by the decoder.
name	str	'aegmm'	Name of the AEGMM model.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor]

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

• Type: Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor]

Decoder

Inherits from: Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

Decoder(self, decoder_net: keras.src.models.model.Model, name: str = 'decoder') -> None

Name	Type	Default	Description
decoder_net	keras.src.models.model.Model		Layers for the decoder wrapped in a tf.keras.Sequential class.
name	str	'decoder'	Name of decoder.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> tensorflow.python.framework.tensor.Tensor

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

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

DecoderLSTM

Inherits from: Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

DecoderLSTM(self, latent_dim: int, output_dim: int, output_activation: str = None, name: str = 'decoder_lstm') -> None

Name	Type	Default	Description
latent_dim	int		Latent dimension.
output_dim	int		Decoder output dimension.
output_activation	Optional[str]	None	Activation used in the Dense output layer.
name	str	'decoder_lstm'	Name of decoder.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor, init_state: List[tensorflow.python.framework.tensor.Tensor]) -> Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor, List[tensorflow.python.framework.tensor.Tensor]]

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor
init_state	List[tensorflow.python.framework.tensor.Tensor]

Returns

• Type: Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor, List[tensorflow.python.framework.tensor.Tensor]]

EncoderAE

Inherits from: Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

EncoderAE(self, encoder_net: keras.src.models.model.Model, name: str = 'encoder_ae') -> None

Name	Type	Default	Description
encoder_net	keras.src.models.model.Model		Layers for the encoder wrapped in a tf.keras.Sequential class.
name	str	'encoder_ae'	Name of encoder.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> tensorflow.python.framework.tensor.Tensor

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

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

EncoderLSTM

Inherits from: Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

EncoderLSTM(self, latent_dim: int, name: str = 'encoder_lstm') -> None

Name	Type	Default	Description
latent_dim	int		Latent dimension. Must be an even number given the bidirectional encoder.
name	str	'encoder_lstm'	Name of encoder.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> Tuple[tensorflow.python.framework.tensor.Tensor, List[tensorflow.python.framework.tensor.Tensor]]

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

• Type: Tuple[tensorflow.python.framework.tensor.Tensor, List[tensorflow.python.framework.tensor.Tensor]]

EncoderVAE

Inherits from: Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

EncoderVAE(self, encoder_net: keras.src.models.model.Model, latent_dim: int, name: str = 'encoder_vae') -> None

Name	Type	Default	Description
encoder_net	keras.src.models.model.Model		Layers for the encoder wrapped in a tf.keras.Sequential class.
latent_dim	int		Dimensionality of the latent space.
name	str	'encoder_vae'	Name of encoder.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor]

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

• Type: Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor]

Sampling

Inherits from: Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Reparametrization trick. Uses (z_mean, z_log_var) to sample the latent vector z.

Methods

call

call(inputs: Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor]) -> tensorflow.python.framework.tensor.Tensor

Sample z.

Name	Type	Default	Description
inputs	Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor]		Tuple with mean and log variance.

Returns

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

Seq2Seq

Inherits from: Model, TensorFlowTrainer, Trainer, Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

Seq2Seq(self, encoder_net: alibi_detect.models.tensorflow.autoencoder.EncoderLSTM, decoder_net: alibi_detect.models.tensorflow.autoencoder.DecoderLSTM, threshold_net: keras.src.models.model.Model, n_features: int, score_fn: Callable = <function squared_difference at 0x117cb00d0>, beta: float = 1.0, name: str = 'seq2seq') -> None

Name	Type	Default	Description
encoder_net	alibi_detect.models.tensorflow.autoencoder.EncoderLSTM		Encoder network.
decoder_net	alibi_detect.models.tensorflow.autoencoder.DecoderLSTM		Decoder network.
threshold_net	keras.src.models.model.Model		Regression network used to estimate threshold.
n_features	int		Number of features.
score_fn	Callable	<function squared_difference at 0x117cb00d0>	Function used for outlier score.
beta	float	1.0	Weight on the threshold estimation loss term.
name	str	'seq2seq'	Name of the seq2seq model.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> tensorflow.python.framework.tensor.Tensor

Forward pass used for teacher-forcing training.

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

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

decode_seq

decode_seq(x: numpy.ndarray) -> Tuple[numpy.ndarray, numpy.ndarray]

Sequence decoding and threshold estimation used for inference.

Name	Type	Default	Description
x	numpy.ndarray

Returns

• Type: Tuple[numpy.ndarray, numpy.ndarray]

VAE

Inherits from: Model, TensorFlowTrainer, Trainer, Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

VAE(self, encoder_net: keras.src.models.model.Model, decoder_net: keras.src.models.model.Model, latent_dim: int, beta: float = 1.0, name: str = 'vae') -> None

Name	Type	Default	Description
encoder_net	keras.src.models.model.Model		Layers for the encoder wrapped in a tf.keras.Sequential class.
decoder_net	keras.src.models.model.Model		Layers for the decoder wrapped in a tf.keras.Sequential class.
latent_dim	int		Dimensionality of the latent space.
beta	float	1.0	Beta parameter for KL-divergence loss term.
name	str	'vae'	Name of VAE model.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> tensorflow.python.framework.tensor.Tensor

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

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

VAEGMM

Inherits from: Model, TensorFlowTrainer, Trainer, Layer, TFLayer, KerasAutoTrackable, AutoTrackable, Trackable, Operation, KerasSaveable

Constructor

VAEGMM(self, encoder_net: keras.src.models.model.Model, decoder_net: keras.src.models.model.Model, gmm_density_net: keras.src.models.model.Model, n_gmm: int, latent_dim: int, recon_features: Callable = <function eucl_cosim_features at 0x280c66430>, beta: float = 1.0, name: str = 'vaegmm') -> None

Name	Type	Default	Description
encoder_net	keras.src.models.model.Model		Layers for the encoder wrapped in a tf.keras.Sequential class.
decoder_net	keras.src.models.model.Model		Layers for the decoder wrapped in a tf.keras.Sequential class.
gmm_density_net	keras.src.models.model.Model		Layers for the GMM network wrapped in a tf.keras.Sequential class.
n_gmm	int		Number of components in GMM.
latent_dim	int		Dimensionality of the latent space.
recon_features	Callable	<function eucl_cosim_features at 0x280c66430>	Function to extract features from the reconstructed instance by the decoder.
beta	float	1.0	Beta parameter for KL-divergence loss term.
name	str	'vaegmm'	Name of the VAEGMM model.

Methods

call

call(x: tensorflow.python.framework.tensor.Tensor) -> Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor]

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor

Returns

• Type: Tuple[tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor, tensorflow.python.framework.tensor.Tensor]

Functions

eucl_cosim_features

eucl_cosim_features(x: tensorflow.python.framework.tensor.Tensor, y: tensorflow.python.framework.tensor.Tensor, max_eucl: float = 100.0) -> tensorflow.python.framework.tensor.Tensor

Compute features extracted from the reconstructed instance using the

relative Euclidean distance and cosine similarity between 2 tensors.

Name	Type	Default	Description
x	tensorflow.python.framework.tensor.Tensor		Tensor used in feature computation.
y	tensorflow.python.framework.tensor.Tensor		Tensor used in feature computation.
max_eucl	float	100.0	Maximum value to clip relative Euclidean distance by.

Returns

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