Neural Network model

This module defines the neural network metamodel.

class besser.BUML.metamodel.nn.neural_network.BatchNormLayer(name: str, num_features: int, dimension: str, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: NormalizationLayer

Represents a type of layer that normalizes inputs within mini-batches to maintain consistent mean and variance, enhancing training speed and stability.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• num_features (int) – The number of channels or features in each input sample.

• dimension (str) – The dimensionality (1D, 2D, or 3D) of the input data to be normalized using batch normalization.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

num_features

The number of channels or features in each input sample.

Type:

int

dimension

The dimensionality (1D, 2D, or 3D) of the input data to be normalized using batch normalization.

Type:

str

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property dimension: str

Get the dimensionality of the input data to be normalized.

Type:

str

property num_features: int

Get the number of channels or features.

Type:

int

class besser.BUML.metamodel.nn.neural_network.CNN(name: str, kernel_dim: List[int], stride_dim: List[int], padding_amount: int = 0, padding_type: str = 'valid', actv_func: str = None, name_module_input: str = None, input_reused: bool = False, permute_in: bool = False, permute_out: bool = False)

Bases: Layer

Represents a layer that is generally used in convolutional neural networks.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• kernel_dim (List[int]) – A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

• stride_dim (List[int]) – A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

• padding_amount (int) – The amount of padding added to the input.

• padding_type (str) – The type of padding applied to the input.

• permute_in (bool) – Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• permute_out (bool) – Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

kernel_dim

A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

Type:

List[int]

stride_dim

A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

Type:

List[int]

padding_amount

The amount of padding added to the input.

Type:

int

padding_type

The type of padding applied to the input.

Type:

str

permute_in

Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

permute_out

Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property kernel_dim: List[int]

Get the list of dimensions of the kernel.

Type:

List[int]

property padding_amount: int

Get the amount of padding added to the input.

Type:

int

property padding_type: str

Get the type of padding applied to the input.

Type:

str

property permute_in: bool

Get whether to permute the dim of the input.

Type:

bool

property permute_out: bool

Get whether to permute the dim of the output.

Type:

bool

property stride_dim: List[int]

Get the list of dimensions of the stride.

Type:

List[int]

class besser.BUML.metamodel.nn.neural_network.Configuration(batch_size: int, epochs: int, learning_rate: float, optimizer: str, loss_function: str, metrics: List[str], weight_decay: float = 0, momentum: float = 0)

Bases: object

Represents a collection of parameters essential for training and evaluating neural networks.

Parameters:

• batch_size (int) – The number of data samples processed in each iteration during training or inference in a neural network.

• epochs (int) – It refers to the number of complete passes through the entire dataset during the training, with each epoch consisting of one iteration through all data samples.

• learning_rate (float) – The step size used to update the model parameters during optimization.

• optimizer (str) – The method or algorithm used to adjust the model parameters iteratively during training to minimize the loss function and improve model performance.

• loss_function (str) – The method used to calculate the difference between predicted and actual values, guiding the model towards better predictions.

• List[str] (metrics) – Quantitative measures used to evaluate the performance of NN models.

• weight_decay (float) – It represents the strength of L2 regularisation applied to the model’s parameters during optimization.

• momentum (float) – It represents a hyperparameter in optimization that helps speed up training by using past gradients to smooth out updates.

batch_size

The number of data samples processed in each iteration during training or inference in a neural network.

Type:

int

epochs

It refers to the number of complete passes through the entire dataset during the training, with each epoch consisting of one iteration through all data samples.

Type:

int

learning_rate

The step size used to update the model parameters during optimization.

Type:

float

optimizer

The method or algorithm used to adjust the model parameters iteratively during training to minimize the loss function and improve model performance.

Type:

str

loss_function

The method used to calculate the difference between predicted and actual values, guiding the model towards better predictions.

Type:

str

metrics List[str]

Quantitative measures used to evaluate the performance of NN models.

weight_decay

It represents the strength of L2 regularisation applied to the model’s parameters during optimization.

Type:

float

momentum

It represents a hyperparameter in optimization that helps speed up training by using past gradients to smooth out updates.

Type:

float

property batch_size: int

Get the number of data samples processed in each iteration during training or inference in a neural network.

Type:

int

property epochs: int

Get the number of complete passes through the entire dataset during the training.

Type:

int

property learning_rate: float

Get the step size used to update the model parameters during optimization.

Type:

float

property loss_function: str

Get the method used to calculate the difference between predicted and actual values, guiding the model towards better predictions.

Type:

str

property metrics: List[str]

Get the measures for evaluating the performance of the model.

Type:

List[str]

property momentum: float

Get the value of the momentum hyperparameter.

Type:

float

property optimizer: str

Get the algorithm used to adjust the model parameters iteratively during training to minimize the loss function.

Type:

str

property weight_decay: float

Get the strength of L2 regularisation applied during optimization.

Type:

float

class besser.BUML.metamodel.nn.neural_network.Conv1D(name: str, kernel_dim: List[int], out_channels: int, stride_dim: List[int] = None, in_channels: int = None, padding_amount: int = 0, padding_type: str = 'valid', actv_func: str = None, name_module_input: str = None, input_reused: bool = False, permute_in: bool = False, permute_out: bool = False)

Bases: ConvolutionalLayer

Represents a type of convolutional layer that applies a 1D convolution.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• kernel_dim (List[int]) – A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

• stride_dim (List[int]) – A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

• in_channels (int) – The number of channels in the input image.

• out_channels (int) – The number of channels produced by the convolution.

• padding_amount (int) – The amount of padding added to the input.

• padding_type (str) – The type of padding applied to the input.

• permute_in (bool) – Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• permute_out (bool) – Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

kernel_dim

Inherited from CNN. A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

Type:

List[int]

stride_dim

Inherited from CNN. A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

Type:

List[int]

in_channels

Inherited from ConvolutionalLayer. It represents the number of channels in the input image.

Type:

int

out_channels

Inherited from ConvolutionalLayer. It represents the number of channels produced by the convolution.

Type:

int

padding_amount

Inherited from CNN. It represents the amount of padding added to the input.

Type:

int

padding_type

Inherited from CNN. It represents the type of padding applied to the input.

Type:

str

permute_in

Inherited from CNN. Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

permute_out

Inherited from CNN. Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property kernel_dim: List[int]

Get the list of dimensions of the kernel.

Type:

List[int]

property stride_dim: List[int]

Get the list of dimensions of the stride.

Type:

List[int]

class besser.BUML.metamodel.nn.neural_network.Conv2D(name: str, kernel_dim: List[int], out_channels: int, stride_dim: List[int] = None, in_channels: int = None, padding_amount: int = 0, padding_type: str = 'valid', actv_func: str = None, name_module_input: str = None, input_reused: bool = False, permute_in: bool = False, permute_out: bool = False)

Bases: ConvolutionalLayer

Represents a type of convolutional layer that applies a 2D convolution.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• kernel_dim (List[int]) – A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

• stride_dim (List[int]) – A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

• in_channels (int) – The number of channels in the input image.

• out_channels (int) – The number of channels produced by the convolution.

• padding_amount (int) – The amount of padding added to the input.

• padding_type (str) – The type of padding applied to the input.

• permute_in (bool) – Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• permute_out (bool) – Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

kernel_dim

Inherited from CNN. A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

Type:

List[int]

stride_dim

Inherited from CNN. A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

Type:

List[int]

in_channels

Inherited from ConvolutionalLayer. It represents the number of channels in the input image.

Type:

int

out_channels

Inherited from ConvolutionalLayer. It represents the number of channels produced by the convolution.

Type:

int

padding_amount

Inherited from CNN. It represents the amount of padding added to the input.

Type:

int

padding_type

Inherited from CNN. It represents the type of padding applied to the input.

Type:

str

permute_in

Inherited from CNN. Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

permute_out

Inherited from CNN. Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property kernel_dim: List[int]

Get the list of dimensions of the kernel.

Type:

List[int]

property stride_dim: List[int]

Get the list of dimensions of the stride.

Type:

List[int]

class besser.BUML.metamodel.nn.neural_network.Conv3D(name: str, kernel_dim: List[int], out_channels: int, stride_dim: List[int] = None, in_channels: int = None, padding_amount: int = 0, padding_type: str = 'valid', actv_func: str = None, name_module_input: str = None, input_reused: bool = False, permute_in: bool = False, permute_out: bool = False)

Bases: ConvolutionalLayer

Represents a type of convolutional layer that applies a 3D convolution.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• kernel_dim (List[int]) – A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

• stride_dim (List[int]) – A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

• in_channels (int) – The number of channels in the input image.

• out_channels (int) – The number of channels produced by the convolution.

• padding_amount (int) – The amount of padding added to the input.

• padding_type (str) – The type of padding applied to the input.

• permute_in (bool) – Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• permute_out (bool) – Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

kernel_dim

Inherited from CNN. A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

Type:

List[int]

stride_dim

Inherited from CNN. A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

Type:

List[int]

in_channels

Inherited from ConvolutionalLayer. It represents the number of channels in the input image.

Type:

int

out_channels

Inherited from ConvolutionalLayer. It represents the number of channels produced by the convolution.

Type:

int

padding_amount

Inherited from CNN. It represents the amount of padding added to the input.

Type:

int

padding_type

Inherited from CNN. It represents the type of padding applied to the input.

Type:

str

permute_in

Inherited from CNN. Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

permute_out

Inherited from CNN. Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property kernel_dim: List[int]

Get the list of dimensions of the kernel.

Type:

List[int]

property stride_dim: List[int]

Get the list of dimensions of the stride.

Type:

List[int]

class besser.BUML.metamodel.nn.neural_network.ConvolutionalLayer(name: str, kernel_dim: List[int], out_channels: int, stride_dim: List[int], in_channels: int = None, padding_amount: int = 0, padding_type: str = 'valid', actv_func: str = None, name_module_input: str = None, input_reused: bool = False, permute_in: bool = False, permute_out: bool = False)

Bases: CNN

Represents a convolutional layer.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• kernel_dim (List[int]) – A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

• stride_dim (List[int]) – A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

• in_channels (int) – The number of channels in the input image.

• out_channels (int) – The number of channels produced by the convolution.

• padding_amount (int) – The amount of padding added to the input.

• padding_type (str) – The type of padding applied to the input.

• permute_in (bool) – Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• permute_out (bool) – Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

kernel_dim

Inherited from CNN. A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

Type:

List[int]

stride_dim

Inherited from CNN. A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

Type:

List[int]

in_channels

The number of channels in the input image.

Type:

int

out_channels

The number of channels produced by the convolution.

Type:

int

padding_amount

Inherited from CNN. The amount of padding added to the input.

Type:

int

padding_type

Inherited from CNN. The type of padding applied to the input.

Type:

str

permute_in

Inherited from CNN. Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

permute_out

Inherited from CNN. Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property in_channels: int

Get the number of channels in the input image.

Type:

int

property out_channels: int

Get the number of channels produced by the convolution.

Type:

int

class besser.BUML.metamodel.nn.neural_network.Dataset(name: str, path_data: str, task_type: str = None, input_format: str = None, image: Image = None, labels: set[Label] = None)

Bases: NamedElement

Represents the collection of data instances used for training or evaluation, where each instance comprises features and corresponding labels.

Parameters:

• name (str) – The name of the dataset.

• path_data (str) – The file path or directory location containing the dataset.

• task_type (str) – The type of prediction task associated with the dataset.

• input_format (str) – The format of the input dataset.

• image (Image) – An image instance that contains the shape desired for the images if input_format parameter is set to ‘images’.

• labels (set[Label]) – The set of labels in the dataset.

name

The name of the dataset.

Type:

str

path_data

The file path or directory location containing the dataset.

Type:

str

task_type

The type of prediction task associated with the dataset.

Type:

str

input_format

The format of the input dataset.

Type:

str

image

An image instance that contains the shape desired for the images if input_format parameter is set to ‘images’.

Type:

Image

labels

The set of labels in the dataset.

Type:

set[Label]

__input_format: str

__task_type: str

_abc_impl = <_abc._abc_data object>

add_image(image: Image)

Image: Add the dimensions of the image.

add_label(label: Label)

Label: add a label to the set of labels.

property image: Image

Get the dimensions of the images.

Type:

Image

property input_format: str

Get the format of the input dataset.

Type:

str

property labels: set[Label]

Get the set of labels.

Type:

set[Label]

property path_data: str

Get the directory location containing the dataset.

Type:

str

property task_type: str

Get the type of prediction task associated with the dataset.

Type:

str

class besser.BUML.metamodel.nn.neural_network.DropoutLayer(name: str, rate: float, name_module_input: str = None, input_reused: bool = False)

Bases: LayerModifier

Represents a type of layer that randomly sets a fraction of input units to zero during training to prevent overfitting and improve generalization.

Parameters:

• name (str) – The name of the layer.

• rate (float) – It represents a float between 0 and 1. Fraction of the input units to drop.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

rate

It represents a float between 0 and 1. Fraction of the input units to drop.

Type:

float

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property rate: float

Get the fraction of the input units to drop.

Type:

float

class besser.BUML.metamodel.nn.neural_network.EmbeddingLayer(name: str, num_embeddings: int, embedding_dim: int, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: GeneralLayer

Represents a layer that learns dense vector representations of the input data.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• num_embeddings (int) – The size of the dictionary of embeddings.

• embedding_dim (int) – The size of each embedding vector.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

num_embeddings

The size of the dictionary of embeddings.

Type:

int

embedding_dim

The size of each embedding vector.

Type:

int

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property embedding_dim: int

Get the size of each embedding vector.

Type:

int

property num_embeddings: int

Get the size of the dictionary of embeddings.

Type:

int

class besser.BUML.metamodel.nn.neural_network.Feature(name: str)

Bases: NamedElement

A feature is a measurable property or characteristic of an object used to represent and describe it within a dataset.

Parameters:

name (str) – The name of the feature.

name

The name of the feature.

Type:

str

_abc_impl = <_abc._abc_data object>

class besser.BUML.metamodel.nn.neural_network.FlattenLayer(name: str, start_dim: int = 1, end_dim: int = -1, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: GeneralLayer

Represents a layer that flattens a contiguous range of dims into a tensor.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• start_dim (int) – The first dimension to flatten.

• end_dim (int) – The last dim to flatten.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

start_dim

The first dimension to flatten.

Type:

int

end_dim

The last dim to flatten.

Type:

int

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property end_dim: int

Get the last dimension to flatten.

Type:

int

property start_dim: int

Get the first dimension to flatten.

Type:

int

class besser.BUML.metamodel.nn.neural_network.GRULayer(name: str, hidden_size: int, return_type: str = 'full', input_size: int = None, bidirectional: bool = False, dropout: float = 0.0, batch_first: bool = True, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: RNN

Represents a Gated Recurrent Unit layer.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• input_size (int) – It represents the dimensionality of the input features.

• hidden_size (int) – It represents the number of units in the hidden state, which captures the network’s internal representation of the input sequence.

• bidirectional (bool) – Whether the layer is bidirectional or not.

• dropout (float) – If non-zero, it introduces a Dropout layer on the outputs of the GRU sub layers except the last one.

• batch_first (bool) – If True, the input and output tensors are provided as (batch, seq, feature) instead of (seq, batch, feature). Only relevant to PyTorch.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

• return_type (str) – Whether to return the hidden states, the last output in the output sequence or the full sequence.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

input_size

Inherited from RNN. It represents the dimensionality of the input features.

Type:

int

hidden_size

Inherited from RNN. It represents the number of units in the hidden state, which captures the network’s internal representation of the input sequence.

Type:

int

bidirectional

Inherited from RNN. Whether the layer is bidirectional or not.

Type:

bool

dropout

Inherited from RNN. If non-zero, it introduces a Dropout layer on the outputs of the GRU sub layers except the last one.

Type:

float

batch_first

Inherited from RNN. If True, the input and output tensors are provided as (batch, seq, feature) instead of (seq, batch, feature). Only relevant to PyTorch.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

return_type

Inherited from RNN. Whether to return the hidden states, the last output in the output sequence or the full sequence.

Type:

str

_abc_impl = <_abc._abc_data object>

class besser.BUML.metamodel.nn.neural_network.GeneralLayer(name: str, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: Layer

Represents a layer designed to handle general operations and transformations.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

class besser.BUML.metamodel.nn.neural_network.Image(shape: List[int] = None, normalize: bool = False)

Bases: Feature

Image represents features designed for handling data with spatial characteristics, typically including attributes such as height and width.

Parameters:

• shape (list[int]) – The shape of the image in the form [height, width, channels]. Default to [256, 256].

• normalize (bool) – If true, the images will be normalized to zero mean and unit standard deviation.

shape

The shape of the image in the form [height, width, channels]. Default to [256, 256].

Type:

list[int]

normalize

If true, the images will be normalized to zero mean and unit standard deviation.

Type:

bool

_abc_impl = <_abc._abc_data object>

property normalize: bool

If true, the images will be normalized to zero mean and unit standard deviation.

Type:

bool

property shape: List[int]

Get the shape of the image.

Type:

List[int]

class besser.BUML.metamodel.nn.neural_network.LSTMLayer(name: str, hidden_size: int, return_type: str = 'full', input_size: int = None, bidirectional: bool = False, dropout: float = 0.0, batch_first: bool = True, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: RNN

Represents a Long Short-Term Memory layer.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• input_size (int) – It represents the dimensionality of the input features.

• hidden_size (int) – It represents the number of units in the hidden state, which captures the network’s internal representation of the input sequence.

• bidirectional (bool) – Whether the layer is bidirectional or not.

• dropout (float) – If non-zero, it introduces a Dropout layer on the outputs of the LSTM sub layers except the last one.

• batch_first (bool) – If True, the input and output tensors are provided as (batch, seq, feature) instead of (seq, batch, feature). Only relevant to PyTorch.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

• return_type (str) – Whether to return the hidden states, the last output in the output sequence or the full sequence.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

input_size

Inherited from RNN. It represents the dimensionality of the input features.

Type:

int

hidden_size

Inherited from RNN. It represents the number of units in the hidden state, which captures the network’s internal representation of the input sequence.

Type:

int

bidirectional

Inherited from RNN. Whether the layer is bidirectional or not.

Type:

bool

dropout

Inherited from RNN. If non-zero, it introduces a Dropout layer on the outputs of the LSTM sub layers except the last one.

Type:

float

batch_first

Inherited from RNN. If True, the input and output tensors are provided as (batch, seq, feature) instead of (seq, batch, feature). Only relevant to PyTorch.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

return_type

Inherited from RNN. Whether to return the hidden states, the last output in the output sequence or the full sequence.

Type:

str

_abc_impl = <_abc._abc_data object>

class besser.BUML.metamodel.nn.neural_network.Label(col_name: str, label_name: str = None)

Bases: object

A label is a value assigned to an observation, representing the target variable for prediction.

Parameters:

• col_name (str) – The name of the column containing the labels.

• label_name (str) – The name of a label in the dataset. If the prediction task is regression, it can be omitted.

col_name

The name of the column containing the labels.

Type:

str

label_name

The name of a label in the dataset. If the prediction task is regression, it can be omitted.

Type:

str

property col_name: str

Get the name of the column containing the labels.

Type:

str

property label_name: str

Get the name of the label.

Type:

str

class besser.BUML.metamodel.nn.neural_network.Layer(name: str, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: NamedElement

This class represents a layer of the neural network. It encapsulates attributes such as the name of the layer and the activation function.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

The name of the layer.

Type:

str

actv_func

The type of the activation function.

Type:

str

name_module_input

The name of the layer from which the inputs originate.

Type:

str

input_reused

Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property actv_func: str

Get the actv_func.

Type:

str

property input_reused: bool

Get whether the input to this layer is reused as input to another layer.

Type:

bool

property name_module_input: str

Get the name of the layer from which the inputs originate.

Type:

str

class besser.BUML.metamodel.nn.neural_network.LayerModifier(name: str, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: Layer

Represents a type of layer that applies transformations or

adjustments to other layers, enhancing their behavior or performance.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

class besser.BUML.metamodel.nn.neural_network.LayerNormLayer(name: str, normalized_shape: List[int], actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: NormalizationLayer

Represents a type of layer that normalizes the inputs across the features of a single data sample, rather than across the batch, to stabilize and accelerate training by reducing internal covariate shift.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• normalized_shape (List[int]) – A list refering to the dimensions or axis indices over which layer normalization is applied, specifying which parts of the tensor are normalized.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

normalized_shape

A list refering to the dimensions or axis indices over which layer normalization is applied, specifying which parts of the tensor are normalized.

Type:

List[int]

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property normalized_shape: List[int]

Get the list containing the dimensions or axis indices over which layer normalization is applied.

Type:

List[int]

class besser.BUML.metamodel.nn.neural_network.LinearLayer(name: str, out_features: int, in_features: int = None, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: GeneralLayer

Represents a densely-connected NN layer that applies a linear transformation to the input data.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• in_features (int) – It represents the size of each input sample.

• out_features (int) – It represents the size of each output sample.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

in_features

It represents the size of each input sample.

Type:

int

out_features

It represents the size of each output sample.

Type:

int

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property in_features: int

Get the size of the input sample.

Type:

int

property out_features: int

Get the size of the output sample.

Type:

int

class besser.BUML.metamodel.nn.neural_network.NN(name: str, configuration: Configuration = None, train_data: Dataset = None, test_data: Dataset = None)

Bases: BehaviorImplementation

It is a subclass of the BehaviorImplementation class and comprises the fundamental properties and behaviors of a neural network model.

Parameters:

• name (str) – The name of the neural network model.

• configuration (Configuration) – The parameters related to the NN training and evaluation.

• train_data (Dataset) – The dataset used to train the NN model.

• test_data (Dataset) – The dataset used to evaluate the NN model.

name

The name of the neural network model.

Type:

str

configuration

The parameters related to the NN training and evaluation.

Type:

Configuration

train_data

The dataset used to train the NN model.

Type:

Dataset

test_data

The dataset used to evaluate the NN model.

Type:

Dataset

__layers: List[Layer]

__modules: List[NN | Layer | TensorOp]

__sub_nns: List[NN]

__tensor_ops: List[TensorOp]

_abc_impl = <_abc._abc_data object>

_collect_nn_warnings(warnings: list)

_module_names() → set

Names of every module declared in this NN (layers, tensor_ops, sub_nns).

_validate_dataset_consistency(warnings: list)

Surface mismatches between training and test datasets that usually indicate user error.

_validate_first_module_entry_point(errors: list)

_validate_module_input_references(errors: list)

_validate_module_names(errors: list, warnings: list)

Reject names that aren’t valid Python identifiers; warn (matching NamedElement.name’s warn-not-error stance) on Python keywords.

_validate_module_uniqueness(errors: list)

_validate_numerical_bounds(errors: list)

Reject non-positive sizes/rates that would crash the trainer at runtime.

_validate_sub_nn_acyclic(errors: list) → bool

Detect cycles in the sub-NN graph rooted at this NN. Returns True if a cycle was found.

_validate_sub_nns_recursive(errors: list, warnings: list, _visited: set)

_validate_tensor_op_references(errors: list)

add_configuration(configuration: Configuration) → Self

Self: Add the configuration to the NN model.

add_layer(layer: Layer) → Self

Self: Add a layer to the NN model.

add_sub_nn(sub_nn: NN)

Self: Add a subnn to the NN model.

add_tensor_op(tensor_op: TensorOp) → Self

Self: Add a tensor Op to the NN model.

add_test_data(test_data: Dataset) → Self

Self: Add the test dataset to the NN model.

add_train_data(train_data: Dataset) → Self

Self: Add the training dataset to the NN model.

property configuration: Configuration

Get the parameters related to the NN training and evaluation.

Type:

Configuration

property layers: List[Layer]

Get the list of layers.

Type:

List[Layer]

property modules: List[NN | Layer | TensorOp]

Get the modules list of the main model.

Type:

List[Union[NN, Layer, TensorOp]]

property sub_nns: List[NN]

Get the sub NN models list of the main model.

Type:

List[NN]

property tensor_ops: List[TensorOp]

Get the list of tensor Ops.

Type:

List[TensorOp]

property test_data: Dataset

Get the dataset used to evaluate the NN model.

Type:

Dataset

property train_data: Dataset

Get the dataset used to train the NN model.

Type:

Dataset

validate(raise_exception: bool = True, _visited: set = None) → dict

Validate the neural network model.

Checks performed:

• Module names are unique within this NN scope.

• Layer name_module_input references resolve to a module defined in the same NN.

• TensorOp layers_of_tensors string entries resolve to a module defined in the same NN.

• The first module in the sequence does not declare an input dependency (it is the entry point).

• Sub-NNs are acyclic (no NN directly or transitively contains itself).

• Each sub-NN is itself valid (recursive validation).

• Warnings are emitted for empty NNs and for missing configuration on a top-level NN that has training data.

Parameters:

• raise_exception – If True, raises ValueError when errors are found. Warnings never raise.

• _visited – Internal — tracks NN instances already validated to stop infinite recursion on cyclic sub-NN graphs.

Returns:

{"success": bool, "errors": list[str], "warnings": list[str]}

Return type:

dict

class besser.BUML.metamodel.nn.neural_network.NormalizationLayer(name: str, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: LayerModifier

Represents a type of layer that applies normalization techniques.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

class besser.BUML.metamodel.nn.neural_network.PoolingLayer(name: str, pooling_type: str, dimension: str, kernel_dim: List[int] = None, stride_dim: List[int] = None, padding_amount: int = 0, padding_type: str = 'valid', output_dim: List[int] = None, actv_func: str = None, name_module_input: str = None, input_reused: bool = False, permute_in: bool = False, permute_out: bool = False)

Bases: CNN

Represents a type of layer that performs a pooling operation.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• dimension (str) – The dimensionality (1D, 2D, or 3D) of the pooling operation.

• kernel_dim (List[int]) – A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

• stride_dim (List[int]) – A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

• padding_amount (int) – The amount of padding added to the input.

• padding_type (str) – The type of padding applied to the input.

• pooling_type (str) – The type of pooling. Either average or max.

• dimension – The dimensionality of the pooling. Either 1D, 2D or 3D.

• output_dim (List[int]) – The output dimensions of the adaptive pooling operation. Only relevant for adaptive pooling.

• permute_in (bool) – Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• permute_out (bool) – Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

dimension

The dimensionality (1D, 2D, or 3D) of the pooling operation.

Type:

str

kernel_dim

Inherited from CNN. A list containing the dimensions of the convolving or pooling kernel (i.e., [depth, height, width]).

Type:

List[int]

stride_dim

Inherited from CNN. A list containing the dimensions of the stride of the convolution or pooling (i.e., [depth, height, width]).

Type:

List[int]

padding_amount

Inherited from CNN. It represents the amount of padding added to the input.

Type:

int

padding_type

Inherited from CNN. It represents the type of padding applied to the input.

Type:

str

pooling_type

The type of pooling. Either average or max.

Type:

str

dimension

The dimensionality of the pooling. Either 1D, 2D or 3D.

Type:

str

output_dim

The output dimensions of the adaptive pooling operation. Only relevant for adaptive pooling.

Type:

List[int]

permute_in

Inherited from CNN. Whether the dimensions of the input need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

permute_out

Inherited from CNN. Whether the dimensions of the output need to be permuted. Relevant for PyTorch. It is used to make PyTorch model equivalent to TensorFlow model.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

_abc_impl = <_abc._abc_data object>

property dimension: str

Get the dimensionality of the pooling.

Type:

str

property kernel_dim: List[int]

Get the list of dimensions of the kernel.

Type:

List[int]

property output_dim: List[int]

Get the output dimensions of the adaptive pooling.

Type:

List[int]

property pooling_type: str

Get the type of pooling applied.

Type:

str

property stride_dim: List[int]

Get the list of dimensions of the stride.

Type:

List[int]

class besser.BUML.metamodel.nn.neural_network.RNN(name: str, hidden_size: int, return_type: str = 'full', input_size: int = None, bidirectional: bool = False, dropout: float = 0.0, batch_first: bool = True, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: Layer

Represents a type of layer used in recurrent neural networks (RNN) for processing sequential data by using memory from previous steps to inform current outputs.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• input_size (int) – It represents the dimensionality of the input features.

• hidden_size (int) – It represents the number of units in the hidden state, which captures the network’s internal representation of the input sequence.

• bidirectional (bool) – Whether the layer is bidirectional or not.

• dropout (float) – If non-zero, it introduces a Dropout layer on the outputs of the current sub layers except the last one.

• batch_first (bool) – If True, the input and output tensors are provided as (batch, seq, feature) instead of (seq, batch, feature). Only relevant to PyTorch.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

• return_type (str) – Whether to return the hidden states, the last output in the output sequence or the full sequence.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

input_size

It represents the dimensionality of the input features.

Type:

int

hidden_size

It represents the number of units in the hidden state, which captures the network’s internal representation of the input sequence.

Type:

int

bidirectional

Whether the layer is bidirectional or not.

Type:

bool

dropout

If non-zero, it introduces a Dropout layer on the outputs of the current sub layers except the last one.

Type:

float

batch_first

If True, the input and output tensors are provided as (batch, seq, feature) instead of (seq, batch, feature). Only relevant to PyTorch.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

return_type

Whether to return the hidden states, the last output in the output sequence or the full sequence.

Type:

str

_abc_impl = <_abc._abc_data object>

property batch_first: bool

Get whether the input and output tensors are provided as (batch, seq, feature).

Type:

bool

property bidirectional: bool

Get whether the layer is bidirectional or not.

Type:

bool

property dropout: float

Get the dropout ratio of the layer.

Type:

float

property hidden_size: int

Get the number of units in the hidden state.

Type:

int

property input_size: int

Get the dimensionality of the input features of the layer.

Type:

int

property return_type: str

Whether to return the hidden states, the last output in the output sequence or the full sequence.

Type:

str

class besser.BUML.metamodel.nn.neural_network.SimpleRNNLayer(name: str, hidden_size: int, return_type: str = 'full', input_size: int = None, bidirectional: bool = False, dropout: float = 0.0, batch_first: bool = True, actv_func: str = None, name_module_input: str = None, input_reused: bool = False)

Bases: RNN

Represents a fully-connected RNN layer where the output is to be fed back as the new input.

Parameters:

• name (str) – The name of the layer.

• actv_func (str) – The type of the activation function.

• input_size (int) – It represents the dimensionality of the input features.

• hidden_size (int) – It represents the number of units in the hidden state, which captures the network’s internal representation of the input sequence.

• bidirectional (bool) – Whether the layer is bidirectional or not.

• dropout (float) – If non-zero, it introduces a Dropout layer on the outputs of the RNN sub layers except the last one.

• batch_first (bool) – If True, the input and output tensors are provided as (batch, seq, feature) instead of (seq, batch, feature). Only relevant to PyTorch.

• name_module_input (str) – The name of the layer from which the inputs originate.

• input_reused (bool) – Whether the input to this layer is reused as input to another layer.

• return_type (str) – Whether to return the hidden states, the last output in the output sequence or the full sequence.

name

Inherited from Layer. It represents the name of the layer.

Type:

str

actv_func

Inherited from Layer. It represents the type of the activation function.

Type:

str

input_size

Inherited from RNN. It represents the dimensionality of the input features.

Type:

int

hidden_size

Inherited from RNN. It represents the number of units in the hidden state, which captures the network’s internal representation of the input sequence.

Type:

int

bidirectional

Inherited from RNN. Whether the layer is bidirectional or not.

Type:

bool

dropout

Inherited from RNN. If non-zero, it introduces a Dropout layer on the outputs of the RNN sub layers except the last one.

Type:

float

batch_first

Inherited from RNN. If True, the input and output tensors are provided as (batch, seq, feature) instead of (seq, batch, feature). Only relevant to PyTorch.

Type:

bool

name_module_input

Inherited from Layer. The name of the layer from which the inputs originate.

Type:

str

input_reused

Inherited from Layer. Whether the input to this layer is reused as input to another layer.

Type:

bool

return_type

Inherited from RNN. Whether to return the hidden states, the last output in the output sequence or the full sequence.

Type:

str

_abc_impl = <_abc._abc_data object>

class besser.BUML.metamodel.nn.neural_network.Structured(name: str)

Bases: Feature

Represents features organized in a systematic manner, typically with well-defined columns and rows, often found in tabular datasets.

Parameters:

name (str) – The name of the feature.

name

Inherited from Feature. It represents the name of the feature.

Type:

str

_abc_impl = <_abc._abc_data object>

class besser.BUML.metamodel.nn.neural_network.TensorOp(name: str, tns_type: str, concatenate_dim: int = None, layers_of_tensors: List[str | float] = None, reshape_dim: List[int] = None, transpose_dim: List[int] = None, permute_dim: List[int] = None, input_reused: bool = False)

Bases: NamedElement

This class represents a tensor operation. It encapsulates attributes such as the name and the type of the tensor operation.

Parameters:

• name (str) – The name of the tensor operation.

• type (str) – The type of the tensor operation.

• concatenate_dim (int) – The dimension along which the tensors will be concatenated with the cat operation.

• layers_of_tensors (List[Union[str, float]]) – The list that defines the inputs of the tensor op. Elements of the list can be either names of layers from which the tensors originate or scalar values.

• reshape_dim (List[int]) – A list specifying the new shape of the tensor after the reshape operation.

• transpose_dim (List[int]) – A list specifying the transpose dimensions. Only relevant with the transpose operation.

• permute_dim (List[int]) – A list containing the desired ordering of dimensions. Only relevant with the permute operation.

• input_reused (bool) – Whether the input to this tensor op is reused as input to another layer (or tensor op).

name

The name of the tensor operation.

Type:

str

type

The type of the tensor operation.

Type:

str

concatenate_dim

The dimension along which the tensors will be concatenated with the cat operation.

Type:

int

layers_of_tensors

The list that defines the inputs of the tensor op. Elements of the list can be either names of layers from which the tensors originate or scalar values.

Type:

List[Union[str, float]]

reshape_dim

A list specifying the new shape of the tensor after the reshape operation.

Type:

List[int]

transpose_dim

A list specifying the transpose dimensions. Only relevant with the transpose operation.

Type:

List[int]

permute_dim

A list containing the desired ordering of dimensions. Only relevant with the permute operation.

Type:

List[int]

input_reused

Whether the input to this tensor op is reused as input to another layer (or tensor op).

Type:

bool

_abc_impl = <_abc._abc_data object>

property concatenate_dim: int

Get the dimension along which the tensors will be concatenated with the cat operation.

Type:

int

property input_reused: bool

Get whether the input to this layer is reused as input to another layer.

Type:

bool

property layers_of_tensors: List[str | float]

Get the list that defines the inputs of the tensor op. Elements of the list can be either names of layers from which the tensors originate or scalar values.

Type:

List[Union[str, float]]

property permute_dim: List[int]

Get the list containing the desired ordering of dimensions for permute operation.

Type:

List[int]

property reshape_dim: List[int]

Get the list specifying the new shape of the tensor after reshaping with the view operation.

Type:

List[int]

property tns_type: str

Get the type of the tensorOp.

Type:

str

property transpose_dim: List[int]

Get the list specifying the transpose dimensions.

Type:

List[int]