dmae package¶

Submodules¶

dmae.dissimilarities module¶

The dmae.dissimilarities module implements several dissimilarity functions in tensorflow.

dmae.dissimilarities.chebyshev(X, Y)[source]¶

Computes a pairwise Chevyshev distance between two matrices \(\max{(|\mathbf{x}_i-\mathbf{y}_j|)}\).

Parameters

Xarray-like, shape=(batch_size, n_features): Input batch matrix.
Yarray-like, shape=(n_clusters, n_features): Matrix in which each row represents a centroid of a cluster.

Returns

Zarray-like, shape=(batch_size, n_clusters): Pairwise dissimilarity matrix.

dmae.dissimilarities.cosine(X, Y)[source]¶

Computes a pairwise cosine distance between two matrices \(\mathbf{D}_{ij}=(\mathbf{x}_i \cdot \mathbf{y}_j)/(||\mathbf{x}_i|| \cdot ||\mathbf{y}_j||)\).

Parameters

Xarray-like, shape=(batch_size, n_features): Input batch matrix.
Yarray-like, shape=(n_clusters, n_features): Matrix in which each row represents a centroid of a cluster.

Returns

Zarray-like, shape=(batch_size, n_clusters): Pairwise dissimilarity matrix.

dmae.dissimilarities.euclidean(X, Y)[source]¶

Computes a pairwise Euclidean distance between two matrices \(\mathbf{D}_{ij}=||\mathbf{x}_i-\mathbf{y}_j||\).

Parameters

Xarray-like, shape=(batch_size, n_features): Input batch matrix.
Yarray-like, shape=(n_clusters, n_features): Matrix in which each row represents a centroid of a cluster.

Returns

Zarray-like, shape=(batch_size, n_clusters): Pairwise dissimilarity matrix.

dmae.dissimilarities.kullback_leibler(loggit_P, loggit_Q, eps=0.001, normalization='softmax_abs')[source]¶

Kullback Leibler divergence. \(\sum_x P_x \log{P_x}-P_x \log{Q_x}\)

Parameters

loggit_Parray-like, shape=(batch_size, n_features): Input batch matrix of loggits.
loggit_Qarray-like, shape=(n_features, n_features): Matrix in which each row represents the unsigned loggit of a cluster.
eps: float, default=1e-3: Hyperparameter to avoid numerical issues.
normalization: {str, function}, default=”softmax_abs”: Specifies which normalization function is used to transform the data into probabilities. You can specify a custom functon f(X, eps) with the arguments X and eps, or use a predefined function {“softmax_abs”, “softmax_relu”, “squared_sum”, “abs_sum”, “relu_sum”, “identity”}

Returns

Zarray-like, shape=(batch_size, n_clusters): Pairwise dissimilarity matrix.

dmae.dissimilarities.mahalanobis(X, Y, cov)[source]¶

Computes a pairwise Mahalanobis distance \(\mathbf{D}_{ij}=(\mathbf{x}_i-\mathbf{y}_j)^T \Sigma_j (\mathbf{x}_i-\mathbf{y}_j)\).

Parameters

Xarray-like, shape=(batch_size, n_features): Input batch matrix.
Yarray-like, shape=(n_features, n_features): Matrix in which each row represents a centroid of a cluster.
cov: array-like, shape=(n_clusters, n_features, n_features): 3D Tensor with the inverse covariance matrices of all the clusters.

Returns

Zarray-like, shape=(batch_size, n_clusters): Pairwise dissimilarity matrix.

dmae.dissimilarities.manhattan(X, Y)[source]¶

Computes a pairwise Manhattan distance between two matrices \(\mathbf{D}_{ij}=\sum |\mathbf{x}_i|-|\mathbf{y}_j|\).

Parameters

Xarray-like, shape=(batch_size, n_features): Input batch matrix.
Yarray-like, shape=(n_clusters, n_features): Matrix in which each row represents a centroid of a cluster.

Returns

Zarray-like, shape=(batch_size, n_clusters): Pairwise dissimilarity matrix.

dmae.dissimilarities.minkowsky(X, Y, p)[source]¶

Computes a pairwise Minkowsky distance between two matrices \(\mathbf{D}_{ij}=( \sum |\mathbf{x}_i - \mathbf{y}_j|^p)^{1/p}\).

Parameters

Xarray-like, shape=(batch_size, n_features): Input batch matrix.
Yarray-like, shape=(n_clusters, n_features): Matrix in which each row represents a centroid of a cluster.
pfloat: Order of the Minkowsky distance.

Returns

Zarray-like, shape=(batch_size, n_clusters): Pairwise dissimilarity matrix.

dmae.dissimilarities.toroidal_euclidean(X, Y, interval=<tf.Tensor: shape=(2,), dtype=float32, numpy=array([2., 2.], dtype=float32)>)[source]¶

Euclidean dissimilarity that considers circular boundaries.

Parameters

Xarray-like, shape=(batch_size, n_features): Input batch matrix.
Yarray-like, shape=(n_features, n_features): Matrix in which each row represents a centroid of a cluster.
intervalarray-like, default=tf.constant((2.0, 2.0)): Array representing the range on each axis.

Returns

Zarray-like, shape=(batch_size, n_clusters): Pairwise dissimilarity matrix.

dmae.initializers module¶

The dmae.initializers module implements some initializers for DMAE.

class dmae.initializers.InitIdentityCov(X, n_clusters)[source]¶

Bases: tensorflow.python.keras.initializers.initializers_v2.Initializer

A tf.keras initializer to assign identity matrices to the covariance parameters.

Parameters

X: array-like, shape=(n_samples, n_features): Input data.
n_clusters: int: Number of clusters.

Methods

`__call__`(shape, dtype)	Generates identity matrices for the given shape and type.
`from_config`(config)	Instantiates an initializer from a configuration dictionary.
`get_config`()	Returns the configuration of the initializer as a JSON-serializable dict.

class dmae.initializers.InitKMeans(kmeans_model)[source]¶

Bases: tensorflow.python.keras.initializers.initializers_v2.Initializer

A tf.keras initializer to assign the clusters from a sklearn’s KMeans model.

Parameters

kmeans_model: :mod:`sklearn.cluster.KMeans`: Pretrained KMeans model to initialize DMAE.

Methods

`__call__`(shape, dtype)	Converts KMeans centroids into tensors.
`from_config`(config)	Instantiates an initializer from a configuration dictionary.
`get_config`()	Returns the configuration of the initializer as a JSON-serializable dict.

class dmae.initializers.InitKMeansCov(kmeans_model, X, n_clusters)[source]¶

Bases: tensorflow.python.keras.initializers.initializers_v2.Initializer

A tf.keras initializer to compute covariance matrices from K-means.

Parameters

kmeans_model: :mod:`sklearn.cluster.KMeans`: Pretrained KMeans model to initialize DMAE.
X: array-like, shape=(n_samples, n_features): Input data.
n_clusters: int: Number of clusters.

Methods

`__call__`(shape, dtype)	Computes covariance matrices from the KMeans predictions.
`from_config`(config)	Instantiates an initializer from a configuration dictionary.
`get_config`()	Returns the configuration of the initializer as a JSON-serializable dict.

class dmae.initializers.InitPlusPlus(X, n_clusters, dissimilarity=<function euclidean>, iters=100)[source]¶

Bases: tensorflow.python.keras.initializers.initializers_v2.Initializer

A tf.keras initializer based on K-Means++ that allows dissimilarities.

Parameters

X: array-like, shape=(n_samples, n_features): Input data.
n_clusters: int: Number of clusters.
dissimilarity: function, default: :mod:`dmae.dissimilarities.euclidean`: A tensorflow function that computes a paiwise dissimilarity function between a batch of points and the cluster’s parameters.
iters: int, default: 100: Number of interations to run the K-means++ initialization.

Methods

`__call__`(shape, dtype)	Estimates n_clusters using K-means++
`from_config`(config)	Instantiates an initializer from a configuration dictionary.
`get_config`()	Returns the configuration of the initializer as a JSON-serializable dict.

dmae.layers module¶

The dmae.layers module implements the dissimilarity mixture autoencoder (DMAE) layers as tensorflow keras layers.

class dmae.layers.DissimilarityMixtureAutoencoder(*args, **kwargs)[source]¶

Bases: tensorflow.python.keras.engine.base_layer.Layer

A tf.keras layer with the Dissimilarity Mixture Autoencoder (DMAE).

Parameters

alphafloat: Softmax inverse temperature.
n_clustersint: Number of clusters.
dissimilarityfunction, default = dmae.dissimilarities.euclidean: A tensorflow function that computes a pairwise dissimilarity function between a batch of points and the cluster’s parameters.
trainabledict, default = {“centers”: True, “mixers”: True}: Specifies which parameters are trainable.
initializersdict, default = {“centers”: RandomUniform(-1, 1), “mixers”: Constant(1.0)}: Specifies a keras initializer (tf.keras.initializers) for each parameter.
regularizersdict, default = {“centers”: None, “mixers”: None}: Specifies a keras regularizer (tf.keras.regularizers) for each parameter.

Attributes

activity_regularizer: Optional regularizer function for the output of this layer.
compute_dtype: The dtype of the layer’s computations.
dtype: The dtype of the layer weights.
dtype_policy: The dtype policy associated with this layer.
dynamic: Whether the layer is dynamic (eager-only); set in the constructor.
inbound_nodes: Deprecated, do NOT use! Only for compatibility with external Keras.
input: Retrieves the input tensor(s) of a layer.
input_mask: Retrieves the input mask tensor(s) of a layer.
input_shape: Retrieves the input shape(s) of a layer.
input_spec: InputSpec instance(s) describing the input format for this layer.
losses: List of losses added using the add_loss() API.
metrics: List of metrics added using the add_metric() API.
name: Name of the layer (string), set in the constructor.
name_scope: Returns a tf.name_scope instance for this class.
non_trainable_variables
non_trainable_weights: List of all non-trainable weights tracked by this layer.
outbound_nodes: Deprecated, do NOT use! Only for compatibility with external Keras.
output: Retrieves the output tensor(s) of a layer.
output_mask: Retrieves the output mask tensor(s) of a layer.
output_shape: Retrieves the output shape(s) of a layer.
stateful
submodules: Sequence of all sub-modules.
supports_masking: Whether this layer supports computing a mask using compute_mask.
trainable
trainable_variables: Sequence of trainable variables owned by this module and its submodules.
trainable_weights: List of all trainable weights tracked by this layer.
updates
variable_dtype: Alias of Layer.dtype, the dtype of the weights.
variables: Returns the list of all layer variables/weights.
weights: Returns the list of all layer variables/weights.

Methods

`__call__`(args, *kwargs)	Wraps call, applying pre- and post-processing steps.
`add_loss`(losses, **kwargs)	Add loss tensor(s), potentially dependent on layer inputs.
`add_metric`(value[, name])	Adds metric tensor to the layer.
`add_update`(updates[, inputs])	Add update op(s), potentially dependent on layer inputs.
`add_variable`(args, *kwargs)	Deprecated, do NOT use! Alias for add_weight.
`add_weight`([name, shape, dtype, …])	Adds a new variable to the layer.
`apply`(inputs, args, *kwargs)	Deprecated, do NOT use!
`build`(input_shape)	Builds the tensorflow variables.
`call`(x)	Forward pass in DMAE.
`compute_mask`(inputs[, mask])	Computes an output mask tensor.
`compute_output_shape`(input_shape)	Computes the output shape of the layer.
`compute_output_signature`(input_signature)	Compute the output tensor signature of the layer based on the inputs.
`count_params`()	Count the total number of scalars composing the weights.
`from_config`(config)	Creates a layer from its config.
`get_config`()	Returns the config of the layer.
`get_input_at`(node_index)	Retrieves the input tensor(s) of a layer at a given node.
`get_input_mask_at`(node_index)	Retrieves the input mask tensor(s) of a layer at a given node.
`get_input_shape_at`(node_index)	Retrieves the input shape(s) of a layer at a given node.
`get_losses_for`(inputs)	Deprecated, do NOT use!
`get_output_at`(node_index)	Retrieves the output tensor(s) of a layer at a given node.
`get_output_mask_at`(node_index)	Retrieves the output mask tensor(s) of a layer at a given node.
`get_output_shape_at`(node_index)	Retrieves the output shape(s) of a layer at a given node.
`get_updates_for`(inputs)	Deprecated, do NOT use!
`get_weights`()	Returns the current weights of the layer.
`set_weights`(weights)	Sets the weights of the layer, from Numpy arrays.
`with_name_scope`(method)	Decorator to automatically enter the module name scope.

build(input_shape)[source]¶

Builds the tensorflow variables.

Parameters

input_shapetuple: Input tensor shape.

call(x)[source]¶

Forward pass in DMAE.

Parameters

xarray_like: Input tensor.

Returns

mu_tildearray_like: Soft-assigned centroids.
pi_tildearray_like: Soft-assigned mixing coefficients.

class dmae.layers.DissimilarityMixtureAutoencoderCov(*args, **kwargs)[source]¶

Bases: tensorflow.python.keras.engine.base_layer.Layer

A tf.keras layer with the Dissimilarity Mixture Autoencoder (DMAE). This layer includes a covariance parameter for dissimilarities that allow it.

Parameters

alphafloat: Softmax inverse temperature.
n_clustersint: Number of clusters.
dissimilarityfunction, default = dmae.dissimilarities.mahalanobis: A tensorflow function that computes a pairwise dissimilarity function between a batch of points and the cluster’s parameters.
trainabledict, default = {“centers”: True, “cov”: True, mixers”: True}: Specifies which parameters are trainable.
initializersdict, default = {“centers”: RandomUniform(-1, 1), “cov”: RandomUniform(-1, 1)
“mixers”: :mod:`Constant(1.0)`}: Specifies a keras initializer (tf.keras.initializers) for each parameter.
regularizersdict, default = {“centers”: None, “cov”: None, “mixers”: None}: Specifies a keras regularizer (tf.keras.regularizers) for each parameter.

Attributes

activity_regularizer: Optional regularizer function for the output of this layer.
compute_dtype: The dtype of the layer’s computations.
dtype: The dtype of the layer weights.
dtype_policy: The dtype policy associated with this layer.
dynamic: Whether the layer is dynamic (eager-only); set in the constructor.
inbound_nodes: Deprecated, do NOT use! Only for compatibility with external Keras.
input: Retrieves the input tensor(s) of a layer.
input_mask: Retrieves the input mask tensor(s) of a layer.
input_shape: Retrieves the input shape(s) of a layer.
input_spec: InputSpec instance(s) describing the input format for this layer.
losses: List of losses added using the add_loss() API.
metrics: List of metrics added using the add_metric() API.
name: Name of the layer (string), set in the constructor.
name_scope: Returns a tf.name_scope instance for this class.
non_trainable_variables
non_trainable_weights: List of all non-trainable weights tracked by this layer.
outbound_nodes: Deprecated, do NOT use! Only for compatibility with external Keras.
output: Retrieves the output tensor(s) of a layer.
output_mask: Retrieves the output mask tensor(s) of a layer.
output_shape: Retrieves the output shape(s) of a layer.
stateful
submodules: Sequence of all sub-modules.
supports_masking: Whether this layer supports computing a mask using compute_mask.
trainable
trainable_variables: Sequence of trainable variables owned by this module and its submodules.
trainable_weights: List of all trainable weights tracked by this layer.
updates
variable_dtype: Alias of Layer.dtype, the dtype of the weights.
variables: Returns the list of all layer variables/weights.
weights: Returns the list of all layer variables/weights.

Methods

`__call__`(args, *kwargs)	Wraps call, applying pre- and post-processing steps.
`add_loss`(losses, **kwargs)	Add loss tensor(s), potentially dependent on layer inputs.
`add_metric`(value[, name])	Adds metric tensor to the layer.
`add_update`(updates[, inputs])	Add update op(s), potentially dependent on layer inputs.
`add_variable`(args, *kwargs)	Deprecated, do NOT use! Alias for add_weight.
`add_weight`([name, shape, dtype, …])	Adds a new variable to the layer.
`apply`(inputs, args, *kwargs)	Deprecated, do NOT use!
`build`(input_shape)	Builds the tensorflow variables.
`call`(x)	Forward pass in DMAE.
`compute_mask`(inputs[, mask])	Computes an output mask tensor.
`compute_output_shape`(input_shape)	Computes the output shape of the layer.
`compute_output_signature`(input_signature)	Compute the output tensor signature of the layer based on the inputs.
`count_params`()	Count the total number of scalars composing the weights.
`from_config`(config)	Creates a layer from its config.
`get_config`()	Returns the config of the layer.
`get_input_at`(node_index)	Retrieves the input tensor(s) of a layer at a given node.
`get_input_mask_at`(node_index)	Retrieves the input mask tensor(s) of a layer at a given node.
`get_input_shape_at`(node_index)	Retrieves the input shape(s) of a layer at a given node.
`get_losses_for`(inputs)	Deprecated, do NOT use!
`get_output_at`(node_index)	Retrieves the output tensor(s) of a layer at a given node.
`get_output_mask_at`(node_index)	Retrieves the output mask tensor(s) of a layer at a given node.
`get_output_shape_at`(node_index)	Retrieves the output shape(s) of a layer at a given node.
`get_updates_for`(inputs)	Deprecated, do NOT use!
`get_weights`()	Returns the current weights of the layer.
`set_weights`(weights)	Sets the weights of the layer, from Numpy arrays.
`with_name_scope`(method)	Decorator to automatically enter the module name scope.

build(input_shape)[source]¶

Builds the tensorflow variables.

Parameters

input_shapetuple: Input tensor shape.

call(x)[source]¶

Forward pass in DMAE.

Parameters

xarray_like: Input tensor.

Returns

mu_tildearray_like: Soft-assigned centroids.
Cov_hatarray_like: Soft-assigned covariance matrices.
pi_tildearray_like: Soft-assigned mixing coefficients.

class dmae.layers.DissimilarityMixtureEncoder(*args, **kwargs)[source]¶

Bases: tensorflow.python.keras.engine.base_layer.Layer

A tf.keras layer that implements the dissimilarity mixture encoder (DM-Encoder). It computes the soft assignments using a dissimilarity function from dmae.dissimilarities.

Parameters

alphafloat: Softmax inverse temperature.
n_clustersint: Number of clusters.
dissimilarityfunction, default = dmae.dissimilarities.euclidean: A tensorflow function that computes a pairwise dissimilarity function between a batch of points and the cluster’s parameters.
trainabledict, default = {“centers”: True, “mixers”: True}: Specifies which parameters are trainable.
initializersdict, default = {“centers”: RandomUniform(-1, 1), “mixers”: Constant(1.0)}: Specifies a keras initializer (tf.keras.initializers) for each parameter.
regularizersdict, default = {“centers”: None, “mixers”: None}: Specifies a keras regularizer (tf.keras.regularizers) for each parameter.

Attributes

activity_regularizer: Optional regularizer function for the output of this layer.
compute_dtype: The dtype of the layer’s computations.
dtype: The dtype of the layer weights.
dtype_policy: The dtype policy associated with this layer.
dynamic: Whether the layer is dynamic (eager-only); set in the constructor.
inbound_nodes: Deprecated, do NOT use! Only for compatibility with external Keras.
input: Retrieves the input tensor(s) of a layer.
input_mask: Retrieves the input mask tensor(s) of a layer.
input_shape: Retrieves the input shape(s) of a layer.
input_spec: InputSpec instance(s) describing the input format for this layer.
losses: List of losses added using the add_loss() API.
metrics: List of metrics added using the add_metric() API.
name: Name of the layer (string), set in the constructor.
name_scope: Returns a tf.name_scope instance for this class.
non_trainable_variables
non_trainable_weights: List of all non-trainable weights tracked by this layer.
outbound_nodes: Deprecated, do NOT use! Only for compatibility with external Keras.
output: Retrieves the output tensor(s) of a layer.
output_mask: Retrieves the output mask tensor(s) of a layer.
output_shape: Retrieves the output shape(s) of a layer.
stateful
submodules: Sequence of all sub-modules.
supports_masking: Whether this layer supports computing a mask using compute_mask.
trainable
trainable_variables: Sequence of trainable variables owned by this module and its submodules.
trainable_weights: List of all trainable weights tracked by this layer.
updates
variable_dtype: Alias of Layer.dtype, the dtype of the weights.
variables: Returns the list of all layer variables/weights.
weights: Returns the list of all layer variables/weights.

Methods

`__call__`(args, *kwargs)	Wraps call, applying pre- and post-processing steps.
`add_loss`(losses, **kwargs)	Add loss tensor(s), potentially dependent on layer inputs.
`add_metric`(value[, name])	Adds metric tensor to the layer.
`add_update`(updates[, inputs])	Add update op(s), potentially dependent on layer inputs.
`add_variable`(args, *kwargs)	Deprecated, do NOT use! Alias for add_weight.
`add_weight`([name, shape, dtype, …])	Adds a new variable to the layer.
`apply`(inputs, args, *kwargs)	Deprecated, do NOT use!
`build`(input_shape)	Builds the tensorflow variables.
`call`(x)	Forward pass in DM-Encoder.
`compute_mask`(inputs[, mask])	Computes an output mask tensor.
`compute_output_shape`(input_shape)	Computes the output shape of the layer.
`compute_output_signature`(input_signature)	Compute the output tensor signature of the layer based on the inputs.
`count_params`()	Count the total number of scalars composing the weights.
`from_config`(config)	Creates a layer from its config.
`get_config`()	Returns the config of the layer.
`get_input_at`(node_index)	Retrieves the input tensor(s) of a layer at a given node.
`get_input_mask_at`(node_index)	Retrieves the input mask tensor(s) of a layer at a given node.
`get_input_shape_at`(node_index)	Retrieves the input shape(s) of a layer at a given node.
`get_losses_for`(inputs)	Deprecated, do NOT use!
`get_output_at`(node_index)	Retrieves the output tensor(s) of a layer at a given node.
`get_output_mask_at`(node_index)	Retrieves the output mask tensor(s) of a layer at a given node.
`get_output_shape_at`(node_index)	Retrieves the output shape(s) of a layer at a given node.
`get_updates_for`(inputs)	Deprecated, do NOT use!
`get_weights`()	Returns the current weights of the layer.
`set_weights`(weights)	Sets the weights of the layer, from Numpy arrays.
`with_name_scope`(method)	Decorator to automatically enter the module name scope.

build(input_shape)[source]¶

Builds the tensorflow variables.

Parameters

input_shapetuple: Input tensor shape.

call(x)[source]¶

Forward pass in DM-Encoder.

Parameters

xarray_like: Input tensor.

Returns

Sarray_like: Soft assignments.

class dmae.layers.DissimilarityMixtureEncoderCov(*args, **kwargs)[source]¶

Bases: tensorflow.python.keras.engine.base_layer.Layer

A tf.keras layer that implements the dissimilarity mixture encoder (DM-Encoder). It computes the soft assignments using a dissimilarity function from dmae.dissimilarities. This layer includes a covariance parameter for dissimilarities that allow it.

Parameters

alphafloat: Softmax inverse temperature.
n_clustersint: Number of clusters.
dissimilarityfunction, default = dmae.dissimilarities.mahalanobis: A tensorflow function that computes a pairwise dissimilarity function between a batch of points and the cluster’s parameters.
trainabledict, default = {“centers”: True, “cov”: True, mixers”: True}: Specifies which parameters are trainable.
initializersdict, default = {“centers”: RandomUniform(-1, 1), “cov”: RandomUniform(-1, 1)
“mixers”: :mod:`Constant(1.0)`}: Specifies a keras initializer (tf.keras.initializers) for each parameter.
regularizersdict, default = {“centers”: None, “cov”: None, “mixers”: None}: Specifies a keras regularizer (tf.keras.regularizers) for each parameter.

Attributes

activity_regularizer: Optional regularizer function for the output of this layer.
compute_dtype: The dtype of the layer’s computations.
dtype: The dtype of the layer weights.
dtype_policy: The dtype policy associated with this layer.
dynamic: Whether the layer is dynamic (eager-only); set in the constructor.
inbound_nodes: Deprecated, do NOT use! Only for compatibility with external Keras.
input: Retrieves the input tensor(s) of a layer.
input_mask: Retrieves the input mask tensor(s) of a layer.
input_shape: Retrieves the input shape(s) of a layer.
input_spec: InputSpec instance(s) describing the input format for this layer.
losses: List of losses added using the add_loss() API.
metrics: List of metrics added using the add_metric() API.
name: Name of the layer (string), set in the constructor.
name_scope: Returns a tf.name_scope instance for this class.
non_trainable_variables
non_trainable_weights: List of all non-trainable weights tracked by this layer.
outbound_nodes: Deprecated, do NOT use! Only for compatibility with external Keras.
output: Retrieves the output tensor(s) of a layer.
output_mask: Retrieves the output mask tensor(s) of a layer.
output_shape: Retrieves the output shape(s) of a layer.
stateful
submodules: Sequence of all sub-modules.
supports_masking: Whether this layer supports computing a mask using compute_mask.
trainable
trainable_variables: Sequence of trainable variables owned by this module and its submodules.
trainable_weights: List of all trainable weights tracked by this layer.
updates
variable_dtype: Alias of Layer.dtype, the dtype of the weights.
variables: Returns the list of all layer variables/weights.
weights: Returns the list of all layer variables/weights.

Methods

`__call__`(args, *kwargs)	Wraps call, applying pre- and post-processing steps.
`add_loss`(losses, **kwargs)	Add loss tensor(s), potentially dependent on layer inputs.
`add_metric`(value[, name])	Adds metric tensor to the layer.
`add_update`(updates[, inputs])	Add update op(s), potentially dependent on layer inputs.
`add_variable`(args, *kwargs)	Deprecated, do NOT use! Alias for add_weight.
`add_weight`([name, shape, dtype, …])	Adds a new variable to the layer.
`apply`(inputs, args, *kwargs)	Deprecated, do NOT use!
`build`(input_shape)	Builds the tensorflow variables.
`call`(x)	Forward pass in DM-Encoder.
`compute_mask`(inputs[, mask])	Computes an output mask tensor.
`compute_output_shape`(input_shape)	Computes the output shape of the layer.
`compute_output_signature`(input_signature)	Compute the output tensor signature of the layer based on the inputs.
`count_params`()	Count the total number of scalars composing the weights.
`from_config`(config)	Creates a layer from its config.
`get_config`()	Returns the config of the layer.
`get_input_at`(node_index)	Retrieves the input tensor(s) of a layer at a given node.
`get_input_mask_at`(node_index)	Retrieves the input mask tensor(s) of a layer at a given node.
`get_input_shape_at`(node_index)	Retrieves the input shape(s) of a layer at a given node.
`get_losses_for`(inputs)	Deprecated, do NOT use!
`get_output_at`(node_index)	Retrieves the output tensor(s) of a layer at a given node.
`get_output_mask_at`(node_index)	Retrieves the output mask tensor(s) of a layer at a given node.
`get_output_shape_at`(node_index)	Retrieves the output shape(s) of a layer at a given node.
`get_updates_for`(inputs)	Deprecated, do NOT use!
`get_weights`()	Returns the current weights of the layer.
`set_weights`(weights)	Sets the weights of the layer, from Numpy arrays.
`with_name_scope`(method)	Decorator to automatically enter the module name scope.

build(input_shape)[source]¶

Builds the tensorflow variables.

Parameters

input_shapetuple: Input tensor shape.

call(x)[source]¶

Forward pass in DM-Encoder.

Parameters

xarray_like: Input tensor.

Returns

Sarray_like: Soft assignments.

dmae.losses module¶

The dmae.losses module implements several loss functions for each dissimilarity in dmae.dissimilarities.

dmae.losses.chebyshev_loss(X, mu_tilde, pi_tilde, alpha)[source]¶

Computes the Chebyshev loss.

Parameters

X: array-like, shape=(batch_size, n_features): Input batch matrix.
mu_tilde: array-like, shape=(batch_size, n_features): Matrix in which each row represents the assigned mean vector.
pi_tilde: array-like, shape=(batch_size, ): Vector in which each element represents the assigned mixing coefficient.
alpha: float: Softmax inverse temperature.

Returns

loss: float: Computed loss for each sample.

dmae.losses.cosine_loss(X, mu_tilde, pi_tilde, alpha)[source]¶

Computes the cosine loss.

Parameters

X: array-like, shape=(batch_size, n_features): Input batch matrix.
mu_tilde: array-like, shape=(batch_size, n_features): Matrix in which each row represents the assigned mean vector.
pi_tilde: array-like, shape=(batch_size, ): Vector in which each element represents the assigned mixing coefficient.
alpha: float: Softmax inverse temperature.

Returns

loss: array-like, shape=(batch_size, ): Computed loss for each sample.

dmae.losses.euclidean_loss(X, mu_tilde, pi_tilde, alpha)[source]¶

Computes the Euclidean loss.

Parameters

X: array-like, shape=(batch_size, n_features): Input batch matrix.
mu_tilde: array-like, shape=(batch_size, n_features): Matrix in which each row represents the assigned mean vector.
pi_tilde: array-like, shape=(batch_size, ): Vector in which each element represents the assigned mixing coefficient.
alpha: float: Softmax inverse temperature.

Returns

loss: array-like, shape=(batch_size, ): Computed loss for each sample.

dmae.losses.kullback_leibler_loss(loggit_P, loggit_Q_tilde, pi_tilde, alpha, eps=0.001, normalization='softmax_abs')[source]¶

Loss for the Kullback Leibler divergence.

Parameters

loggit_P: array-like, shape=(batch_size, n_features): Input batch loggits (pre-normalization values).
loggit_Q_tilde: array-like, shape=(batch_size, n_features): Cluster loggits (pre-normalization values)
pi_tilde: array-like, shape=(batch_size, ): Vector in which each element represents the assigned mixing coefficient.
alpha: float: Softmax inverse temperature.
normalization: {str, function}, default=”softmax_abs”: Specifies which normalization function is used to transform the data into probabilities. You can specify a custom functon f(X, eps) with the arguments X and eps, or use a predefined function {“softmax_abs”, “softmax_relu”, “squared_sum”, “abs_sum”, “relu_sum”, “identity”}

Returns

loss: float: Computed loss for each batch.

dmae.losses.mahalanobis_loss(X, mu_tilde, Cov_tilde, pi_tilde, alpha)[source]¶

Computes the Mahalanobis loss.

Parameters

X: array-like, shape=(batch_size, n_features): Input batch matrix.
mu_tilde: array-like, shape=(batch_size, n_features): Matrix in which each row represents the assigned mean vector.
Cov_tilde: array-like, shape=(batch_size, n_features, n_features): Tensor with the assigned covariances.
pi_tilde: array-like, shape=(batch_size, ): Vector in which each element represents the assigned mixing coefficient.
alpha: float: Softmax inverse temperature.

Returns

loss: array-like, shape=(batch_size, ): Computed loss for each sample.

dmae.losses.manhattan_loss(X, mu_tilde, pi_tilde, alpha)[source]¶

Computes the Manhattan loss.

Parameters

X: array-like, shape=(batch_size, n_features): Input batch matrix.
mu_tilde: array-like, shape=(batch_size, n_features): Matrix in which each row represents the assigned mean vector.
pi_tilde: array-like, shape=(batch_size, ): Vector in which each element represents the assigned mixing coefficient.
alpha: float: Softmax inverse temperature.

Returns

loss: array-like, shape=(batch_size, ): Computed loss for each sample.

dmae.losses.minkowsky_loss(X, mu_tilde, pi_tilde, alpha, p)[source]¶

Computes the Minkowsky loss.

Parameters

X: array-like, shape=(batch_size, n_features): Input batch matrix.
mu_tilde: array-like, shape=(batch_size, n_features): Matrix in which each row represents the assigned mean vector.
pi_tilde: array-like, shape=(batch_size, ): Vector in which each element represents the assigned mixing coefficient.
alpha: float: Softmax inverse temperature.
p: float: Order of the Minkowsky distance

Returns

loss: array-like, shape=(batch_size, ): Computed loss for each sample.

dmae.losses.toroidal_euclidean_loss(X, mu_tilde, pi_tilde, alpha, interval=<tf.Tensor: shape=(2,), dtype=float32, numpy=array([2., 2.], dtype=float32)>)[source]¶

Loss for the toroidal euclidean dissimilarity.

Parameters

X: array-like, shape=(batch_size, n_features): Input batch matrix.
mu_tilde: array-like, shape=(batch_size, n_features): Matrix in which each row represents the assigned mean vector.
pi_tilde: array-like, shape=(batch_size, ): Vector in which each element represents the assigned mixing coefficient.
alpha: float: Softmax inverse temperature.
intervalarray-like, default=tf.constant((2.0, 2.0)): Array representing the range on each axis.

Returns

loss: float: Computed loss for each batch.

dmae.metrics module¶

The dmae.metrics module implements some evaluation metrics that are used in the paper.

dmae.metrics.unsupervised_classification_accuracy(y_true, y_pred)[source]¶

Scipy-based implementation of the unsupervised classification accuracy.

Parameters

y_true: array-like, shape=(n_samples, ): Array with the Ground truth labels.
y_pred: array-like, shape=(n_samples, ): Array with the predicted labels.

Returns

uacc: float: Unsupervised classification accuracy between y_true and y_pred.

dmae.metrics.zero_norm(preds, thr=1e-07)[source]¶

Numpy implementation of the L0 norm.

Parameters

preds: array-like, shape=(n_samples, n_clusters): Soft-assignments extracted from a DM-Encoder
thr: float: Threshold used to compute the L0 norm.

Returns

L0: float: L0 norm of the soft-assignments.

dmae package¶

Submodules¶

dmae.dissimilarities module¶

dmae.initializers module¶

dmae.layers module¶

dmae.losses module¶

dmae.metrics module¶

Module contents¶